,ID-Ri33 651 DESIGN OF RELATIONAL DATABASE BENCHMARI(S(U) NAVAiL 1/2POSTGRADUATE SCP-OOL MONTEREY CA V C STONE JUN 83

UNCLASSIFIED F/G 9/2 N

Ehhhmmhhhils

1.0i LOo LL-W

ii

mu

1.25

MICROCOPY RESOLUTION TEST CHARTNATIONAL BUREAU OF STANDARDS-1963-A

fNAVAL POSTGRADUATE SCHOOLMonterey, California

cTh

THESISDESIGN OF RELATIONAL DATABASE BENCHMARKS

by

Vincent Courtney Stone

June 1983

LA., Thesis Advisor: David K. Hsiao

Approved for public release; distribution unlimited

.- - ( ,*4 - -- -

SGCUPAv CLMFSIIICAION OF THIS RAGIC (011w Data ftteio_____________

EPOWI'DOCUMNTA~iN PAG INSTRUCTIONS -REPW00CMENATWNPAC BEORZCOMPLETING FORM=PONT~ ~ ~ ~ 11411NII .6RE casp !. 11CIPIRMT1S CATALOG NUIA11111

. TITLE (amt aI9.) S. TYPE OF REPORT & PERIOD COVEREDMaster's Thesis

Design of Relational Database Benchmarks June, 1983S. PERFORMING ORG. REPORT MUMMER

7. AaiTHWA) 6. CONTRACT ON GRANT HUMMER(s)

Vincent Courtney Stone

IL PEFDMNGDANIATION NAME AND ADDRESS 10. PROGRAM ELEMENT. PROJECT. TASKAREA & WORK UNIT NUMESERS

Naval Postgraduate SchoolMonterey, California 93940

It. CONTROLLING OFFICE NAME AND ADDRIESS 12. REPORT DATENaval Postgraduate School June 1983onterey, California 93940 1). Mumma" rOPAGES

____ ___ ____ ___ ___ ____ ___ ___ ____ ___ ___ 11214. 111011TORING AWENCY NAME G AODRESWeI "am Ie CM0001109 Offic IS. SECURITY CLASS. (of thd. mprFI)

UNCLASSIFIED15a. OE9CLASSIFICATION/OiOWNGCRAOING

SCH EDULE

W. DIGTmEUTION STATERIENT (of~ ANSlbefl

Approved for public release; distribution unlimited 5

17. DmSvINIUTION STATEMENT (of Me abooini eewd in A1e00 M it.111f~eame bs XOPeM)

WS SUPPLEMENTASIY NOTES]

benchmarking, database machines

21L AWSYRACT (C.Utee mi cem e it ****o smm! Wi lrdaty Or Slk flwi~)

Performance measurements of a database machine reflect not onlythe processing power of the machine, but also the size andstructure of the database. It is therefore useful to constructdatabases for performance measurements of database machines.Furthermore, it is u-seful to utilize synthetic data, such thatthe volume of the reply can be predicted for a given query and thestructure and attributes of the database can be varied (Continued)

%/W 0102- LP. 014- 6601 1 SECURITY CLASSIICATION4 OF TH4IS PAGE Om0an flee. DItwo'

. ...... ..-. - ...-

i TV CLAMPCATION OF T141 PAGE Mb= Ske 81s

ABSTRACT (Continued) Block # 20

for intended test queries. Conducting measurement studies usinga synthetic database contributes to the generality of the resultswhen different test queries are employed. A parameterized prograuis described herein which can be used to generate various relat-ions for a synthetic database. The experiences in constructingand using the database generator are described. It is suggestedthat given sufficient information on real-world databases thegenerator may be useful for modeling them as well as for creatingdatabases for benchmark tests. *1

Acepqs~lc T' r

*.---'-.

I . ! ." ' -. C'

0 .•102- L..014.6601

ICCUNITY CLAMOPICATION O T11 P"6 eNf m Do& -

Approved for pue.ilc release, distribution unlirnitel.

S

Design of

Relqationai Database Bencnmarcs

by

Vincent Courtney StoneLieutenant Commander, United States NavyB.S., United States Naval Academy, 174 . -

Sublitted in partial fulfillment or therequirements for tne aegree of

.IASTER OF SCIENCE IN COMPUTER SCIENCE

from the

NAVAL POSTGRADUATE SCFOOLJune, 11e-3

Au tho r:---------------

Ap r v d by ----(---Thesis Advisor

Sec o e r I

Chairman, Department or Computer Science

Dean of Informationan y sciences

2

.* a.., *. . . , . .. . .. . .. . . . " 2''

ABSTRACT

Performance measurements of a database machine reflect

not only tfe processinr power of tne macline, cut &aso tne

size an! structure of tne database. It is tterefore useful

to construct databases for performance measurements if iata-

base ma',fnines. Furztermore, it is useful to utilize

syntnetic data, sucn tnat tne volume of tne reply can be

predicted for a riven query and the structure and attributes

of the database can be varied for intended test 4uerips.

Conductina measurement studies usinr a syntnetic daatabae

contributes to the generality of the results wren different

test queries are employed. A parameterized program is de-

scribed nerein wn ict can be used to generate various

relations for a synthetic database. The experiences in con-

structing and using the database generator are descriet It

is sugested tnat given sufficient information on real-worli

databases the renerator may be useful for modelinr tnen as

well as for creatinR databases for oencnmark tests

3

."W ') 'J,' '.' :.' .':,,'',',:-'...:,.'..,,'.-.-'. . - . -

-',.. . . . . . .. . . . . . . . . . . . . ..-.. ... .-. .-... ...-...... ...... .-.-

TABLE OF CONTENTS

I. BENCHMARKS FOR DATABASE MACHINES -I

1. PERFORMANCE MEASUREMENTS -

B. BENCHMARKING . . . . . . .. . . . .. 11

C. VUANTITIES TO B MEASURSD 11

II. BENCHMARKING RELATIONAL DATASASE MACHINS -1

A. THE BENCHMARKING ENVIRONMENT ------------- 13

B. THE ARCHITECTURE OF THE SYSTEM ------ 14

1. Tne Basic Macnine Arcnltecture andVarious Configurations 14

2. The Database Organization-------------- 16

3. The User Interface -------------... 17

III. THE BENCHMARKING APPROACH ----- 2G

A. A MULTI-DIMENSIONAL PRDPLEM ------ - 2

1. Moaeling Problems --- -2

a. DBMS-dependent Proolems .......... 20

b. Database-dependent Protle.ns -..... 2v

2. Measurement Problems ....... 21

B. RESOLVING THE MODELING PROBLEMS -------------- 22

1. DBMS-dependent Proolems ----------------

2. Databoase-dependent Problens --

C. THE SINTHESIZED DATABASE AND WORKLOAD ?4

1. The Use of Synt1esized Data ..... et

2. Types of Syntnesized Data------- ....--

3. General Scnema of Syntnesizea Data Usei - 2t

IV. GENERATING SNTHESIZED DATA --------------------- 29

4

A. A PAR&METERIZED RELATION GENERATOR - 29

1. Capabilities ....... 29

2. The Development 3V

a. The Development Environment 3

b. The Development Pro-ess ------------ 3

c. Design Problems ---------------------- 61

B. A MATRIX OF RELATIONS ----------------- 33

1. Standard Templates ---------------------

2. Flexibility 35

C. THE GENERATING PROCEDURE 45

1. The Generator System ---. ..-

a. The Relation Generator (RG) -

b. The Value-Set Generator (VG) 3

2. The Conversion Problem .............. 3Y

F. Transporting the Relations to thneTest bed . .. . .. .

a. Transporting tne Data to tne Host --- 4V

b. Loading Data Into tfe Bacicend 4- -

V, GENERATING TEST PROGRAMS --------------- ------ 42

A. THE TEST PLAN ------------------------

1. Experiments Involving a Sincie Relation - 42

2. Experiments Involving More Than One Reia-t1 on 46

3. A Flexible Test Plan ------------

B. MEASUREMENT TOOLS ----------------------- 4F

C. OUERY SCRIPTS vs. PROGRAMS --------------- 49

5

D. INTERPRETING THS DATA --------- -

-I. CONCLUS IONS ----.--- -------- -----

A. RESULTS -...... .- - - 52

1. General Results - -5Z

2. Researcn Results --

B. A RELATIONAL PENCHMARLING , ETHODOLO'Y 5

1. Paase One - Measurement Metnods 55

2. Pftase Two - Component Isolation --------- 5

3. Phase Taree - System Response ------------- 1

4. Pnase Four - Verification ---- ---

C. SUMM&RT - ----------- --------------

APPENDIX A IDM System Tables ---------------- 64

APPENDIX B Database Generator Proeram (CMS PascalVS) - 65

APPENDIX C Dataoase Generator Frorram (,1VS PascalYS) - b5

APPENDIX D Valueset Generator Program (MVS PascalVS) - 102

APPENDIX E Valueset Generator Proeram (CMS PascalVS) - l(t

LIST OF REFERENCES ------------------ --------- 109

BIkLIOGRAPHY -------------------

INITIAL DISTRIBUTION LIST i-----------------------------111

6

f, -''.t -'

9.-_ ..... . . . . . . ... .. ,..'....,....

List of Figures

Fi. 1. Tfte IDMi b us Arctiitecture - - - lb

Fig. 2. Tfte IDM/User Interface Diagram -..... 1

Fig. 3. Response Time vs. Size of Returned rata -

7

.- 7:- . .

LIST OF TABLES

Table 1. The Characteristics of a Relation----------------------27

Table 2. The Gneral Relation Template--------------------------34

Table 3. The Schema of an Attribute Record----------------------37

8

ACKNOOLEDGEMENTS

I would lite to acrowlelge tne nelp and support of tne

followi.g people in %me preparation of tr.is thesis: Doris

Mieczzo and tne staff of tne Data Processing Support Center,

lest, Point Mairu, California, CDR T. M. Piposcl, USN, Naval

e Security Group Heaaquarters, Paula R. Strawser, my secona-

reader and overseer ani Dr. Davil K. Hsaio, my advisor.

I would also lise to note tnat tne wore performed in

support of tlis tnesis nas been performed in concert Ofit

tfe tnesis worK of LCDR Curtis :. Ryder, USN, LT Rooert A.

Boglanowicz, ant- LT Mictael D. Croc'ter.

I. BENCHMARKS FOR D&AAaE !&L±!Nj

A. PERFORMANCE MEASUREMENTS

In comparing iatatase management systems 'ri Ss) an

important factor is t.teir performance. One way to coripire

DBMSs is to run specific appiziations under a variety of

systems. Eact system can be 'fine-tunel' to eive t.e Dest

result. An evaluation based on sucn a metnoa is costly ani

time-consuminaR. Often such a method !ray be infeasible. In

many cases, a database for tft1* specific applications Tay

not even exist. As a second metnod, an evaluation couid be

male on tne oasis of performance measurements of existinr

dazatases. This methnod is less costly and less time-consu-

mine. Rowever, the followine questions arise. Is the

existing database sufficient to support intenaed applica-

tions? Are the applications wood for condu tine relative

performance evaluation of different DBMSs?

It is impractical to perform sucn lirect comparison of

DBmSs. Adapting an application to several systems for eval-

uation ourposes is not practical. Evaluation based on

existinR databases is subject tc interpretation error. Tne

Increasing number of DBr-'Ss maces it Imperat4ve that some

method is to be devised to do comparative performance

measuramen ts.

7 7o .7 7 7- 7-. 77.

B, BENCHMARKING

The concept of a stanard for measuring performance is

not new. The standard is usually icnown as a nencr,rarl,

after tle markers used by surveyors in establ sting a common

reference point for their measurements. For exar'jle, Mount

Diablo (a mountain east of San Francisco) is usei as Tne

reference point in surveyine mucn of Nortnern Calit'orn!ia ue

to its long-rane visitility. A meinia for measurnp

similar items in reference to a stanrarl is caliel

henchmaricIng.

Precedents for vencnmarrinR. exist in measurine the per-

formance of computer systems. The Gibsor-Mix metn.o

measures the execution time of a specific set of application

programs for bencnmarcting computer systems. The exnectei

a performance of a system could be computed by cnaracterizinK

the expeztel wor~load as a mix of jobs trom the standara

set.

It is proposed that a set of application programs can ne

devised to measure the performance of DkMSs. Usinp these

bencnmaric measurements, it will te possible to rompare tne

performance of various DBM3Ss. The .easurements can be ana-

lvzed to sureest strenetns and weacnesses of the DtSs.

C. QUANTITIES TO 31 MEASURED

Tne cenerally acceptea performance Intex for a DtS Ib

trte response time. Defining tne response t Ie as tne

11

I,,*4-** ., ... .- -.. - . ...-...

: ' . : ' , '-. - L , ' __. " " ' " " . . . - . - '4. - . . . " ".." • , . " " - . . " .' -

primary performance index is tne scope of tP.is researcz.

fowever, the response time Is based on several factors.

Among tnese factors are tne time to process tne auery, tne

" " time to access the data, t-e time to process data, ani the

time to return tne data. For a DBMS runnine on a mainframe

computer, t!,,e effects of other woriloaa on the response

time must also be considered.

A measurement of tne response time is more sirirl-ant

when measurements of its components are provided. SvYre

simplifyine assumptions may be made. The first su-n assurp-

Lion is that tne rate of accessing data in tne datatase is

constant. The second is thnat tne rate of returning pro-

cessed lata is constant. However, tne time involvel In tne

processing of queries and tne time involved In tne proces-

sine of data may vary ereatly amone ,atabase operations. It

order to record the variance of time among tnE operations,

tests must be devised vnlcn will indicate tnese components

for all supported operations.

This tnesis focuses on measurements of tne response

time. A development of a system to measure coMponents of

tne response time is iscussed. Tne system involves tne

eeneration of a synthetic latabase. Tne system also rea-

sures tq.e Oencftnarted macftine in usine that database.

u'" T . . " - '- : - t - ~ ' - " ' N " '" ' : " . .. .. - ."- ' '." -" " .'. " " -

II. BENCHMARKING RSIL119N &L PAflJAIZ4 tI~a

A. THE BENCHM&RKIN3 ENVIRONMENT

The researcn done in support or tflis tnesis nas been

performed In a complex environment. The complexity Involves

multiple macnines and multiple operatine bystems.

h Relation Generator (RG) of synttetic relations nas

been developed usine Pascai (i.e., Ibm's Pascal/VS) in a

multiuser environment (VM/CMS running on I3M 3063). RG is

usal In a oatcn environment (MVS) on the same macnine. The

o ,relations are zenerated in FBCDIC-cnaracter form. Tney are

transported to a UNIVAC 110e via tape. The EBCDIC riles ar!

tnen loaded onto tae nost (i.e., tae UNIVAC computer) and

translated oy tte host into ASCII files. Tnese ASCII riles

are finally lnaded into a bacceni dataoase macnite .i..,

Britton Lee's IDM 500).

Tae oactend macnine and interface software for tne 11Z

series computers are marKetei by the Amperir Corporation or

C.atswortn, California, as tae RDM 1100. A.ditional mea-

surements can be made ty typassine tne part or tne queryprocessor ttat provides terminal support. Tnis Is accom-

plisned by co'i unlcatine tirectly witft the query processor

via compiled language statements (i.e., COBOL). Tnis does

not completely by ass tfte ouery processor, because tne query

language is Interpreted and cannot be precompiled. However,

13

[o,- -

i

tne results snow tnat query processing does not represent a

sinificant portion of the response time if tne nost ,orx-

load Is light. The terinal nandler represents also a small

portion of the response time. Therefore, the only alvantape

to the use of compiled programs is tne option of runrirng, tne

process as a bactground job.

B. THE ARCHITECTURE OF TF.E S1STEM

The arcnitecture of the system encompasses two major

areas. The first of these areas is the internal arc.itec-

ture of tne IDM 500. The second area is tne nost systei

software, i.e., the user Interface wnich runs on the host.

1. The Bas ic Macn__ & o..nZ!An:

The IDM 50 is made up or several modules connectel

to a common niet-speed bus (See Figure 1). The database

. process5r is a 6-mnz, ZiloR Z-8fe series ml'roprocessor

whicn performs the DBMS functions. The coding for tne

microprocessor is written largely In tne C programming lan-

guage, along wit some assembly lanruaae routines. It

comprises about 330 1t-bytes of machine code. r. optional

module, the iatama e a1i1e:11.a1: Improves tne system per±.'r-

Mance by i. plementlne in nln-speed, special-purpose

hardware some of the DBMS functions normally performed ov

tne lataoase processor.

II1 4-"

Expandable

Database Cache.,Host Cch

Interface Processor Memory

HIGH SPEED BUS

Database

.4 Accelerator

(optional) Controller

The

Database

Figure 1 - The .IDM Bus Architecture

::; 15

- -

The V C..rX is composed of C4i-Dit dynaric ..

cnips. Tne tasic configuration (at tne beginninz o+" tie

tests) includel one-half iregatyte of memory. Up to six

''megatytes of memory can te supported. During tn=e testing

period, configurations of ote and two meoatytes nave also

been used.

One t o four I i sg -ontrollers may De installe-I.

!a:n controller supports up to tour six-nunore-megatytE,

nard ~ ~ ~ ~ _1. dmk. A~ lj~ay te ins talleet to far-ili-

tate bacting up and loading data.

Two standard nost interfaces are availatle. A

ISEE-4E8 byte-wide parallel interface is availaole for con-

nection to mainframes and minicomputers. A second interfa'e

can be used to provide multiple RS-232 serial ports to

microcomputers. A special byte/word interface for CoM-

- iunication witn UNIVAC tost computers is supplied by the

Amperif Corporation.

2. The Database Organization

The IDM1 50C software supports the relational lata-

base model. Data is stored on tne disk in two Ic-ica

levels. These levels are tne system datatase and tne user

databases. At tne top level, tne system database -ontins

five system tables and. tnirteen aatatase tables. T.ie systei

tables contain information on nardware confi.uratlon, aata-

bases and current usace. The tlirteen iattas-e tatles

comprise tne data dictionary. Tney are usei to

16

14W:

store information about relations, attributes, users, ani

security. A list of tne system tables and tfle iatabase

tables is riven In Appendix A.

Altfout. access to tne system database is require!

for tie creation of a user latabase, an existing user data-

base can be accessed directly, i.e., witnout zoin tnroufn

the system database. Each user database nas totr. latacase

tables and user tables. Tae database tao±es are stored

-ithin t.qe user database and may be accessed in tne same

manner as user tables.

The basic unit of lisK access is a 2K-byte blocK.

When a database is created, a space allocation in viocts may

be requested. Tnis allocation may be increased if neces-

sary. Both system tabtles and database taDles are used by

tte system to compute physical addresses.

3. Tne Utr Interface

Tne user Interface is accessed by invocing an prn-

cess on tne !ost. This process is an interactive query

processor. Tne auerv processor parses tne user's queries

written in the Relational Query Lanruae (P.L). RCL is

Amperif's Implementation of Britto.-Lee's Intelligent Query

Language (IOL). Alternativeiy, queries may te sucmitted to

tne query processor from a compiled COBOL or FORTRAN pro-

gram. Submitting a compiled program as a batcn job, tne

user may bypass tre query processor's terminal tanaier.

° - - % '° , . % o o . .- . .. . % 1?

However, the bitcn job still depends on the query pro-estor

for parsing of tne query.

The, Relational Query Language (RQL) Drovies c.e.-

tions and facilities similar to tnose avaliatle on

relational DBMSs currently running on mainframe ,omputers

and larger minicomputers. RQL also allows 4ueries to oe

pre-parsed and stored within a database. Tnese storea ccr'-

manis limit toe time required in toe nost for parslng an

reduce tne ti-re required in the bactend tor the tatabase-

table lookup. Additional information on RQL ray be found in

[1, 2 and 3J.

Communication with the IDM is via a system process,

RDMIO. RDMIO supervises communications between user pro'es-

ses running on the host and the hardware interface to tfne

IDM (See Figure 2). Up to ten users may access tne Fr!

simultaneously from a single UNIVAC host.

11

S -'W.'_ " ",", " " " **ti'<* ". ...

I - . '4. 4 4-- .•

TerminalBatchUser

-C,0

RQL I(Terminal COBOLInterface) lInterfacel

(Parsing, Query

RQL$ IGeneration, Output

jFormatting)

RDMIO

FIgure 2 -The 1DM/User Interface

* 19

Ill. THE BENCEMARLING APPROACH

A. A MULTI-DIMENSIONAL PROBLEM

Creatinp a tencamarring system poses a pronlem witl

several dimensions. The problem can be trowcen down into two

* major areas. Tnese areas are moaelinj, and measurement.

1. o A eenl Probems

Tne modeling problems can be cateizorized as DbMS-

dependent and database-dependent. Te DBMS-dependent

modeling problems are related to DBMS schema and syntax.

The database-dependent problems are related to tne clara-

teristics of tae database and tne application to be Tcdelea.

a. DEMS-lependent Problems

Tte three widely .nown database models are the

lierarcnical, tne networx, and tte relational. It ras been

shown that databases and applications based on one of these

molels can be translated to any otter model. However, tnere

is no accepted basis for meaninrful comparisons of tneir

performance measurement. As a first step, tests nave been

performed in support for estabcisning such a basis for DPMQs

lavine t.he same underlyinw model, specifically tne relation-

al Model.

b. Data base-depentent Problems

The datatase-dependent problems are represevta-

tive of existine databases and the appltations wl.ln ire

2f0

i_• , • . " . , , *, - •,* --. . -• - . -. . - . , .- , . - + • . . *" ,-+ .. , - . .

used on them. Existine databases vary In tne coMIlexity arl

in the efficiency in which t.ey nave been implementea.

These varieties are partly due to thne pnysical data tnat are

represented in thne database and partly due to tne Prorram-

mers4 abilities to construct tne database. Ad'itionally,

the applications which use these latavases also model tre

pnysical data retresented as well as tne informatiorn re-

quired of tre latabase. Thus, coth existine iatatases ana

applications must be modeled. The key to an effective and

meneral model is creating one waich represents common cna-

rateristics. The characteristics of latabases and

applications must be carwfully studied prior to thne design

or a reneral and effective model. The contrastino nature of

existing databases and taelr applications present an ex-

tremely complex modeling problem.

2. .M.._S ,'_. } r9_.L.~

DBMS bencinmari measurements, as a standard, may also

represent a comparison of DBMS performance. This standard

may be eitner absolute or relative. Absolute measuremerts

assume a fixed standard. Relative measurements may provice

rancings within a group of DBMSs. The 'easurement ort tne

response time for relative rancing is our roal.

Experiments must be constructe. carefully and the

environment must oe controlled to provide useable, accurate

measurements. For example, in pertforminet resear,-n for trIs

21

Sn . n .'

tnesis it nas been noticed. thnat tne load on tne nost cEn

sienificantly affect tne response time as seen by t re user.

Similarly, tne response time is teavily affected oy tne time

reuuired to return the data to the user at the screen.

These effects must be minimized in order to obtair mpisure-

ments wnicn more accurately reflect tne performance o0' tne

bazKend database macnine. Resolution ot measurerent zro-

blems is discussei in Section V.B.

B. RESOLVING THE MODELING PROELEMS

Altnougn the modeling problems cannot re Plirrinate,

steps can be taten to minimize tne errors introdu.cel t tne

eoeeling process.1. DSMS-,tndenj Proqep

Two assumptions can be made to minimize tne Dn"S-

lependent modeling errors. Tnese assumptions concerr the

format of tne data and the operations used to access the

data.

The first assumption is that all relations ire

stored in tnird normal form (NF). Tne use of 3N.F mlinimizes

the possibility of inconsistent lata. 4Vile real rta es

-lo not use 3NF, this fact ioesn't discourage cur assumDticn.'A

The benchmari is deslwned to provide a measurement ot' r'Livs

performance. It is not intended to taie into consiaeration

4Z

* - . .* .. .. . . .

the abilities of tnose persons wno will design tne datacases

(altbouan ease or use may be a consideration in so-re

instances), for they may not understand tne tneory cf ;5NfF.

Tne second assumption to De made is tnat tne query

languages used by tne DBMSs are logically equivalent. Ai-

tnougfn differences in syntax do exist, tney generally an not

affect tne breadth of available operations. T,erefore, i

common set of queries can be implemented in tne IBMs"

individual syntaxes and provide tne iaentieal lo-irai r-

sult. Any variations to tnis should be noted with bencnn'arK

results. The basic set of experiments include selections,

projections, loins, updates, insertions and 1elilons. Ad-

ditionally experiments snoula be performed wnicn test tne

performance of any peculiar or powerful operations which a

DBMS may nave in addition to tne standard set.

2. Database-de~ende fg aobl

Tne eliiination of aataoase-dependent modeling pri-

bl.ms involves two fundamental areas. The first of tnese

areas is tne Reneration of a syntnetic database. The -ene-

ration of sucn a datacase allows tne use of data wti'n is

cenerally representative of existin databases, tut nft

spezifically representative of any one. The lesion o" tne

syntnetic database's cnaracteristics snould De broad. This

" - ensures tnat it can be adaptea to realistically measure tne

performance of a database with its own !narateristi-s.

These cnaracteristics inclute tne sizes of tne relations (in

23

"a

tne number of tuples and tuple lengtzn) in tne datavase a.:

the lengtt of a tuple relative to blocK size of tne storape

medium.

The second area involving database aeuer.etrcy

involves ztne applications running on tne database. A syn-

tnetic wortload is required for tne same reabcns .s for tne

- syntnetic database. Tne design of tne syntnetic wcr loa!

S.1OU11 be broad enourn to provide enourf, results to te atie

* - to fully simulate different applications. Tae wor.rload is

designed witn two major considerations. Tne first consi--

ration is support of tne basic relational operitions

liscussea previously. An additional conslaeration tares

into account tne varyinw access patterns of eu istir.e data-

oases. For example, a given application may repeatedly

retrieve only one tuple at a time. Anotner will retrieve

many in one operation. An iTportant cfaracteristi: is t ."

locality of tne data retrieved by oDerations. Tnis Cnarac-

teristic may produce different levels of performance wit

lifferent indexing metnods.

C. THE STNTHESIZED DATAASE AND '4ORKLOAD

1- Mae Uz of Svn1R.1pZ2. P.AIIn tetermining a set of bencnmar measurements, it

is necessary to obtain tne set wnicn car. te usea on a wiae

range of DBMSs. It is also irrportant tnat tnis set ioes not

favor any DBS or class of JOBMSs.

24

S,

Two approaches coull nave teen taKen in ottainlino

measurements. One approach dould te to perform tests on

existine databases. The other approact is to do measure-

ments on a synthetic database. The latter allows thie

createst flexibility in performing operations on tne ,ata-

base. This is because tne scnema of a real aetabase rr ht

not provile a suitable structure ror Derforminr a test o f

some operations. Tae schema of a syntnetic database, on Tne

other nand vinimizes any bias resulting from designing tne

tests around a particular database.

The research for this tnesis is performed in con-

junction with evaluation of relational database mac.nines.

However, the installation nas no relational datatasec.

Therefore, any tests on t.e DBMMS would nave to be .et#.rrmea

on either a synthetic iataase or a database ^onverted f.!o0

another model. Since tne use of synthetic dtatat.ases sup-

ports a more general approach in bencn marring, the cnoice

has been .tade to generate sucn databases for bencnmarwicniz

tests.

2. T 2T yta11lps

SyntnesizeI data snoull nave one major cnaracterist-

ic. The types of data snould be oroaa enouin to test tne

supported DB 1S operations of ilfferent types of r ielcs

(i.e., values). Yor example, in tne research performed for

tnis thesis, tne first two attribute values of eacn relation

25

nave tne same numeric value. However, the first attribute

value is stored is an Integer and the second as a -r.ir!zter

string. One set of tests selects tuples based on tn te-

ger values; a second set of tests selects tne sare tu-jes

based on character values. Response times may oe af'e:te!

by the processin, differences related to tne data types.

Adaitional lifferences may result from tne time requirei tc

format tne iata for output.3. General _nema of Synjnes& zen.1 ig _

Tne syntnesized data used for tnls tnesis has four

sets of relations. Each set nas several relations with

different numbers of tuples. Each relation in a set nas tte

same attributes. The attritutes are similar amonr tne f u.r

sets, differing only in number and leawt l in order to pro-

vide a range of tuple lengths. Tatle I snows the range of

tuple ciaracteristics.

The relations are stored in several dataoases. Two

latabases ire used for testine sinale-relation operations.

The first latabase contairs all of the relations usea In

sincle relation testine. The second database 7ontains rqla-

tions wnose tuDles are of 100 bytes and 200 ,ytes. Tnis

database uses compressei fields for strings (i.e., traiLinz

blanis are dropped) . Several dataoase5 are used ti trovirte

relations for testing join (,perations. For testinR, it is

desirable to spreai tne join operations over tne two alszs

in tne system. A full implementation cf tais cresirable

26

.. %...... ".. ".-'.-...v.....

4q~ Table 1. Tfte Relation CnaracteristIcs

Tuple Lenetas : 100, 204, 1000, 20LA Bvtes

Relation Loical Sizes : 500, 1000, 2500, 59,, 1VO0 Tuples

Relation Pnysical Sizes : 50 X11obytes to 20 megabyteS

Attributes : 14 ( for 100 tyte tuples,, ?4 (forozner tuple lengtns)

Attribute Types : Sequential Integer, Random Inteer,Coliatei Alphanumeric, Bloc s Sets

27

database place-ent is not possible, because tne s torai;-=

allocation algoritams prevent us from controilng over tne

storage location of specific relations.

°-1

-:4

S. . . . * * *' . .

S. . . . . . V .'t'' - ~**S...S. * * . . . . . . S S .2 X ql

.1 Q

A. A PARAMETERIZED RELATION GENERATOR

The Relation Generator (RG) is a parameterized p.oeram

for reneratine relations for a latabase. PG prorrpt5 t.,e

user concerning tne cnaracteristirs of a relaticn. Firs!

ttie user is instructet to enter t.te relation na. Aind slzZ

(i.e., tne nvmber or tuples). Tnen, tne program requests

'ata about eac.t attribute. T.e aata equesteia includes

attribute nane, value type (i.e., integer, strlne, etc.) an4

distribution of m~e attribute values. Tne relations gene-

rated are stored in ASCII flies to simplify transfer between

systems.

RG contains routines to generate sequentlal nurroers,

random numbers (eltner uniquely or nonuniqueiy), and cna-

racter strings in collated oraer (See Appendix B). The user

may also specify a file whicn contains a set of values 4or

an attribute to be used in generating attribute values.

Tnis set is called a 'value-set' and tne file Is called a

-value-etj tile- It is produced oy tne utility proi'ram.

Value-set Generator (aeSiribed below). Tile actual ranre or

K values from tne file to be used for an attribute Is callei

K the attribute's iomain. The 'user spe'tl-'es tne nuircer o!'

values from tne value-set to be includea in t.e attr1rute's

i.'.L -. °,

V .7. 77- --7 -- '

"omaln. It is not necessary that tle domain contain ali tnc.

values in the value-set. RG requires tne user to defir.e tne

distribution of the attribute values. Tne distritutior. is

either in discrete blocis or random or both. Ai iscrete

distribution in wnicn tne attribute values are ranaomly dis-

tributed may be created by sortina a relation containin,

discrete oloc_ on a random number attritute.

2. Trio DIafljn

a. The Development Environment

RG is written in IBM Pascal/VS, runnine under

the VM/CMS operating system. Vm/CMS is an interactive,

3multiuser operating system. Beacause of operating syste,

limitations, RG has been converted to a MVS (tatch) process.

Standard Pascal syntax aas been utilized as mucn as possi-

ble. Pascal/VS extensions to tne laneuage nave teen used.

Additionally, some of tie file aescriptor Inforr ation is

specific to thne operating systems.

b. The Development Process

Tne first step in thne development of tne system

is thne draftinw of a modular frameworK. Persons are tnen

assiened to develop tae different moiules of tnp prcgram.

The different modules include tne main program, tne main

generator module and tte Individual value-type generatcr

molules. The Individual modules produce speciflc types of

values for tne attributes.

,30

, ...

The system nas been leveloped usine modern o:t-

ware engineering tecnniques. The different modules nive

been debugges separately. Program harnesses, wniCn contain

no logic except to invoke a procedure, nave beer usea tc

test procedures and subpro.eiures. Module stubs. wnicn

simulate tne actions usually performed by procetures, nave

been used in place of tne procedures to test tne main pro-

gram ard tne main generator module. Once de~uggel, tne

modules nave been intewrated witn tne main program.

The responsibility for generating relations aa_

been assigned to one person. Additional levelopment ot tne

!. system involved several items in addition to aebu 'ein-. e

utility to wenerate value-set files has also been treates.

Thus, the other members of the team nave been freed to wor:c

on other phases o tne project.

c. Design Problems

Two major problems have been enrounterea In the

preparation of RG. The first problem is tne sIZe of tne

relations to be renerated. In the orriinal I l.esitn. all

of the lince! lists of attribute values reside in tne prima-

ry memory simultaneously. Tne size of thne largest relatior.

that has teen generated is twenty megabytes. Tnis requires

twenty megabytes of tne virtual memory space just to store

the contents of the lists. Additional space woull te re-

quired for the program and the overtead associated witnr lined lists (i.e., pointers to memory locations). Tnis

% .e .1 (..e.,

iL

exceeds the virtual memlory space available to a single user

under VM/CMS.

This problem nas been partially solvel cy acces-

.-, sine sequential files as a Substitute for the linced i1sts.

Therefore only one list of attribute values at a tire Is

stored in the primary memory. However, a lilrea list 't

some of the loncer attributes eneratel reouires over two

megabytes of memory just for tne data, Oitnout conslerir.-;

tie space required for pointers.

Tne seconi problem concerns the transportation

of the files of renerated relations to artotner system.

Under the VM/CMS system at the Naval Postgraduate Scnool,

eacn user is allowed a limited amount of file space. Tnis

amount Is mucn too small to hold most of tne relatiors gene-

rated. Additional space Is available on a temporarv (i.e.,

one-day) basis. Also important Is t.e fact that wnile

VM/CMS files can be offloaded to tape, tney are storea on

tape in a non-standard format. There is no utility propram,

to transfer VM/C.MS files to tape in standarf format. T ere

is also no utility program to exchange files between the

tapes of VM/CcS format and tne tapes of M7S format.

It is apparent tnat VM/CMS is not the Ideal er-

vironment in whirch to run the system. Therefore. it n a

been necessary to convert the system to run In tno V en-

vironment. The MVS system writes tapes in tre stanlara

32

;- m % ,% % , " , -"• •" "•". -. ..

format. It also allows the user to have a mucr. lreer

virtual memory space. In retrospect, it maces sense ti'

levelop the system in an interactive system (i.e.. "Yv/ vS .

• ii Fast turnaround contributes to faster program levelopment,

and tne interactive environment mares debugging easier.

B. A MATRIX OF RELATIONS

Tne relations generated b y RG are aesionsd to _upptirt

experiments over a ranre of relation sizes and .naracteris-

tics. These sizes and characteristics are seiecte, to allow

maximum flexibility in pursuing experilents witn a minimal

number of relations In tne test database. Tne parameters

discussed below are those of tne relations priducea in sup-

port of the benctmarrinr.

1. Stagard TeM _ a

Ail of the relations are cnaracterized by tae same

reneral template. This template is shown in Table 2. Four

specific templates are derived from tne general one. These

templates correspond to tne four tuple lengtns uset for

testine (i.e., 100 bytes, 200 bytes, 100, ,ytes and :?VkZ

bytes). Eacn template Is used to generate tne relations of

various sizes (bo - 10,( tuples). Tnus most of tre tests

can be run on many relatilns by cnaniine only tne relation

namie (or the values of the range variable) in the queries.

.433

-. 7 -.- 7 6 -.- - ;.

Table 2. _eej:a_ Ra_ lon T D2_aIe

ley - a sequential number to be stored as an ir.-tecer field

*.Irror - a sequential number (same as iev) to restorel as a onaraf'ter strine

Random - a ranlom number to be stored as an integerfield

Ranlom tIni4ue - a uni4ue random number to be stored as arinzeRer field

Colilate - a cnaracter string to be stored in aipnabetic or-

der

Letter - a random aipnabetical letter

sets - blocrs of values from value-set files.

not usel in some templates

multiple attributes dependine on tne tupie lenietn

34

0

The relations are tesiRnel to provide fieXiJility in

testing. Ideally tte tests to be performed are snowL- te'oreWlesirnine the relations. However, thie results rrcnr. sore or

tne tests may suggest a need for additional tests wnicn aalre

not been previously considered. Accorlintly, tne relations

are lesianed to allow the desizn of additional tests witl:out

generating ore relations.

C. THE GENERATING PROCEDURE

Tme generating procedure consists of three pnases. Tne

frst phase consists of designing experiments and tne rela-

tions to be used In those experiments. &rter the relations

nave been desirned, they must be created and transpcrtel to

tne testing environment.

Generating relations is a simple process. First VG, is

usel to generate any necessary value-set files. Then, RG is

used to generate relations. RG nas been expanded to pronuce

.1 a description file. This file contains the attribute na-'es

and cnaracteristics of tne attrioute values in tne relation.

The description lists both tne format of the aeneraterl file

and tne format of tne relation as it is to be stirei in tne

database.

1. The_ GenerorA _zyl.ep

The Renerator system consists ot' t#o major proprams,

the Relation Generator (RG) and tne Value-set Generator

.3

v C * <'5

(2). Otner programs and debugging aids may Oe nacessary,

dependine on the environament(s) in wnicr, the syster is

implemented.

a. The 'Relation Generator (RG)

RG creates a relation file based on Input from

the user, It consists of four types of modules: tne rrial..

program, tne main generator module, tne individual .enerator

molules, and tae collating module.

The Main Module - Tne main RG module contains

very simple logic. R% prompts tne user for tine cnaracteri-

stics of the relation being generated. First, tne name and

size (in tuples) of tne relation is requested. Tnen. tne

user is asred to letermine tne cnaracteristics of tte first

attribute. Tne attribute cnaracteristics are collected in

an attribute record (See Table 3). After tne module obtains

tne necessarv attribute cnaracteristics, it invoies tne main

generator module.

The main renerator module, as explned tintne next

section produpes linked lists of attribute values ani re-

turns to tae main RG module. RG tnen Invoices tne collate

module which is detailed in tine sequel. Tne collatp !odlile

produces tuples by concatenatinR sets of attrioute values.

After tne relation nas been generated, tne ter is giver tne

option of reneratin, another relation or enline tne process.

The Main Generator Module - The rai!. ceneratcr

module is invoted to produce each set of attribute values.

°36

Tatile. 3 The ScneMA 2t' an A11rj1tlt Recori

Attribute Name - assioned attribute nameAttribute Type - data type 0or attribute vaiuesString Lengtn - used for string typesLover Bound - first sequential Integer ana. lo-

ver bound for random inteeerstipper Bound - upper bound on random integersGenerate Mode - data-type distritutiorValue Set Name - value-set file nameRelative Proportions - discrete distribution SDecit'1-

cationSeed - random Integers

.37

i 3?

................

The characteristics of an attribute are passei to the -oc,±e

in an attribute record. Using tnis record, tne main -o:ule

invoKes one of several individual generator moauie., :e=e--

tine on tne -naract eristics of tne attriDt_. Tre

inlividual generator module produces a lintea list cf attri-

bute values witn tne aesirea. type and distritution, ani

returns the list to tne main generator moiule. T e main

eenerator Todule opens a sequential rile, writes tn ttri-

bute values Into the file, closes tie file, ani returns to

tne main RG nodule. There are therefore several sucn :lies,

%cnown as attribute files.

Collate Module - Tne collate module acts as a

collator. It paysically concatenates strings of attrioute

values to form a tuple. It is invoked to assimilate all tne

attribute values in tne attribute files into a file of tne

relation. In'formation describing tne attributes is passed

to tne collator as an array of attribute re'orls. Tre

collator first opens tne relation file, and all tne attri-

oute files. Tne relation is generated a tuple at a time.

One attribute value fron each file is read. The values are

concatenated to produce a tuple. The tuple is tnen written

to t..e relation file. Tne collator repeats tnis process

until all tne tuoles nave been produced.

b. T!e Value-set Generator (VG)

Tne Value-set Generator (VG) is a simple utility

for setting up value-set flies for RG. VG asks tor tne name

. . -

mw,,

and size (i.e., the number of values) of the valre-set fiie

to be created. The values are entered indivilually In:

stored as strings in a random-access file for use ty RG.

2. TP-e 22jj2 krq.

Converting the program to run in tae batcn envi-

ron.ment involves several tasis. Tnese are tne conversion of

interactive programs to batch Drozrams, tne submission o.

jobs to tne batcn system, and developmert of' the adaitionai

statements requirel to use of tne batch file system. Al-

tnougn tne programs nad already been debugged in tne VM/1nS

environment, extensive dtebureine has been necessary after

conversion to MVS.

Conversion of proerams from VM/CMS to MVS is

not a simple process. A virtual card decir is createe In a

7M/CMS file wnicn contains tne source decK, tne Input data=

and the file data required Dy tne MVS system. Tnis flie is

submitted to tne batcn aueue. Tne Input for RG (i.e., tne

user's replies) are in the carl lecK witp the proerar.

Altnougn it is not necessary, tne source code

w.icn genei-ated tne instructions to tne user for tne input

has been reToved for tne MVS versions. Tne VM/CMS version

.as been modified to create a file waicn contains tne user's

responses to tne program's prompts.

Differences between tne batcn and ir.trartive

* systems caused the dif i ulty in proeram conversion. Tne

W * ,- ,'. .'.% ' " ",. - ' .-. -" . . . - .", " . . . - . • " . • " , . " . -

batch syste.m, MVS, requires mucn more in tte way of file

parameter specifications, and is much less forrivine wnen

error conditions exist. There are some error corcitiors

wnich the user can not foresee. For example, the system nay

initially allocate space for a relation file on a volume

wnicn does not have enough free space to cover secondary

allocations. 'then tnis happens the proerair is aborted.

However, it is not possible for the user to spe:i"y 3 pirtl-

.* cular diskr (i.e., one with sufficient space) for file

storage. For the two largest relation files (fifteen and

twenty meeabytes), it has been necessary to write eacn of

the relations into two separate files on the oatcn system.

.The two files were then combinel when they loaded into tne

latabase.

3. Trans .r ting the Reaions _2_qe Tjj_ _pe

a. Transportine the Data to the East

The transportation of tne relations to tne host

is a two-step process. The 'irst step is tne transfer of

the relation files from tne MYS secondary storage to tape.

A system utility is used to accomplish this. Tho tapes are

then transporte to the host, tne UNIVAC ioi, and a similar

utility prorram is used to load lata into the nost seroniary

storage. The host utility proRram translates t.e EBCDInC

tape files into ASCII ,,sE files.

b. Loadine Data Into the Bacicend

The relations are loaded into tne baciend uslnz

40

a vendor-supplied utility called a translator. This utility

prompts the user for inrormation about tne source file, tne

target database, and the target relation

The translator utility may be run interactively

or with file input. The database into waicn tne relation is

to be loaded must already exist. The relation intc waicn

data is loaded may or may not exist. Database name, no t

file name, and relation narre must be suppilie. Aaditional-

ly, for each attribute tne attribute name, length of source

(in ASCII cbaracters), and type of value to be storec In tr.e

database must be supplied.

4,

-°.

.° .

V. GENERATING TEST PROGRAMS

A. THE TEST PLAN

The general test plan calls for several aifferenz types

of experiments. Among tnese are experiments involving only

one relation (i.e., selections and projections) ana expert-

rments involving more taan one tatabase (i.e., joins).

1. E~x]er12.e a Sin le Relation

The selection and projection expert-rents are le-

signed to measure tne system's performance in retrieving

data from a single relation. The response times measured

are the sum of four variables: the time to process a query,

tne time to access tne data, tne time to process the data,

ani tne time to return the lata. The time to process tre

query is defined as tne time to parse the q uery. By care-

fully constructing sets of experiments, these varlar.les can

be estimated.

Since the time to process a query is so small, it

may be ignored or combined with overnead for most expert-

ments. For experiments wf.ere it is sianitficant, tne

4uery-proccessing time is minimized to prevent it from domi-

nating tne time measurement. resulting in a loss of

precision. The RDM 11?e allows tne parse tree of a juery to

be stored in tne database. This capatility allows tne

replaiement of the processing time, wnicn is lepenlent or

42

-.1

tne host, witn the data access time, which Is mepencent only

on the DacKend. The adcditional data access time is tnfe tire

to access tne command In storaee. This is tne same for

all stored commands.

The larRest variables are tne easiest to measure

with Drecision. Thererore, they are ,measured trst ana tren

eliminated to measure tne smaller variables.

The largest variables are Iiiety to ne those repre-

senting tne time to access, process and return data. Tnese

can be measured with simple retrieve commands. A time

measurement of a retrieve which returns a-1 the attrioute

values of the tuples in a relation Includes tne times of all

of thne four variables. However, a time measurement using an.

arcreeate function (e.r., count, wnich returnb a sinwle

count of tne tuples meeting the qualifications of tne auery)

eliminates the time to return tte lata. Thus this function

can be used effectively to measure the time to access ani

.orocess the data (tuples), i.e., two of tne four variables.

Furtner, an assumption is made ttat for simple iorm-

mands thne processor can process data at a rate wnich is

faster than the rate that data can oe broucnt into tne

* emory for processing. Tnis allows t.e processing time to

be irnored. Therefore, tne measurements reduce to a measure

of t e access time.

43

Vd *~~s~... 7 .

Havine quantified the lareer variables, the time t3

process data may be Investigated. It has been assumec tnat

the processine time is not significant for simple cerm-ands.

However, if te commands are made more complex, tnen t.e

processing time Is expected to increase. 'itn a sutticient-

ly complex command wlich Involves a small lata-access time,

tne processing time may tecome si'nificant. Tnerefore,

experiments are conducted whlich inimize data a-cess Dut

vary in complexity. It is of interest to deterrmine wnen or

if the Drocessinr time becomes measureable and si nificant.

It is expected that projections operations will

increase the processing time. Therefore, several tests are

appropriate for testine projections. The first set of tests

measure the effect of projections on tne processing time.

The second set chects to see if the processini time Is

affected vy the type(s) of attribute values projectea (i.e.,

integer, strine). Tne tnird set of tests measures tne

performance of a projection on all of t.te attrinutes versus

a simple 'retrieve all' command.

After tue time basic variables cave been estimatel,

other perf'ormance factors are investigated. Tne use ,'

Indices can reduce acc.ss time. Py reiucing tne amourt of"

lata brounrt into the memory, tne processing time is al so

reiucei. However, the processing time will be tr.reased lue

to Index aczess and searcn. Therefore, for some relations,

tne use of indiCes may inc rease the resporse tiime. Ini.exir,

44tm s fidcsmy f(e~ fersonetm.Il~il

requires a specific set of tests to measure its per!orrn-ie

in various situations. Tne use of different types of ir.-

-ies (i.e., ¢lustered, non-clustered, multiple .eys, et'.

must also be investigatei. An expected factor in Inex

performance is thne ratio of tne index size (in tloczs or

storage) to that of the relation.

Strine r-ompression (removal or" trailini spaces is a

factor which can affect tqe processing time, tne acc.ss time

ani the return time. The use of compression can reiice

bloci storage Iramaticaly. This, in turn, redu-es trie

access time. However, it may require more time to process a

compressed strine versus a non--ompressed one, if pro-cssinh

requires expansion of tne compressed attribute. It expan-

sion is not required for processine, then thne nost may nave

to expand it for proper formatting. How expensive (in time)

is tnis? Does this nompensate for tne retu'tion in tote

responnse time resulting from returning a smaller amount of

data (the compressed strine) to toe host?.

Otter performance factors may be examined eitner

inliviiually or witnin other test pro'erures. An exarple or

tis is tne use of different types of ittrifutes (i.e..

integer versus string). A complete series of tests can r

leveloped to test this issue in letail. However, it is also

appropriate to investigate tnis area in conjunction wltn

processing time and projections.

45

2. E7Rer1ens nog.la ng More Than One RelatioZ

Operations involving more tnan one relation (i.e.,

joins) are affected by tfte same time variables as t tobi

involving only a sincie relation. Initial testing snoui

involve only two relations.

It is expectea thtat the access time will become

lominant for join operations. Ttis is because tne same lata

may have to be accessed repeatedly. Memory size nas a.

effect on tne aiount of accessing required in a Join opera-

tion. If memory size is larre enourn to allow totn

relations to be accessed once and le t in tne Temory, thnen

thne processing time may become slniricant. In t!is rtirm-

stance ootn tfe access time ani tne processin, time are

expected to increase proportionally to tne relatior. siTe.

The unknown factor is the rate at which tne pro-essing time

increases. However, it may be that neither relition is

small enou.n to be held in tne memory for processing. In

this case much accessiln must be performed. It may also be

of interest to examine join performance between tnese two

extremes.

Tne join snould be aesigne. to tare acvantage oif any

size differential between trhe two relations. If tfle smaller

relation can be completely neld in tne memory, tnen it can

be ac-essei once anc brounft into tne memory. Tte lireer

relation can also be accessed just once as it is nrouent

into tne menory as a strean. If, on tne otner nand, tne

'

'.-.'~~~~~~~~..w .-.- " .". " .....-. - -...-........ .. ... . .

larger relation is brought into tne memory, it Tust te

brourht into tne memory a portion at a time. Tne smaller

relation may nave to be reaccessed for ea' n portion o!' tne

larger relation.

It is important to examine the performance of joins

both with and witnout selection. In per'orminr tnese tests,

the stratery of the operations snould te examinec -areruiy.

The selection snoull be performed before t.e actual join

operation to minimize the volume or data bein joilnea.

Another area of interest is the effect of index

usare on joins. Performance here Is expected to Improve as

inlicated by the sinele relation Index experiments. Fowever

the specific results may suggest tne efficiency with wnicn

the join operation has been implemented.

If inequality joins nave been Implementea, perfor-

mance testing snould be conluctel using tnem. If tney nave

not been implemented, it may be valuable to Xnow If, and

with what lifficulty, tney can be simulated.

Having experimented tne join operations InvolvIne

two relations, experiments operations snould te conqucted

using larger numbers of relations in one join operation. Ey

Investigating tne performance on multiple join reiations, it

may be possible to isolate a flxed overneid for all t.e

initial joins.

4?7

3. & Ie Plan

A gereral test plan snoula e leveloped te.fcre any

of the experiments are designed. It sould be ,iexicie to• ..

enable testing to follow different patns of ciscovery. It

is expected tnat tne results of some experiments may sugest

otter experiments. Time must e alloted for tr.e expansion

of any test set.

Fowever, it rust also ensure tnat tne a 5't'icient

range of data is ootained. The tests must cover tne univer-

sal operations (i.e., those expected of any DBMS). Amonir

the universal operatiors, Known bottleneci's and creaipoints

are specifically tested. It snould also investiz-ate an.-

specific strengtts, weaknesses or idiosyncrasies o. tfne

DBMS.

B. MEASUREMENT TOOLS

The response-time ,measurements in znese experiments were

taken from the Daciend-machine clocK. This ciocic has a

resolution of 1/50 second and an accuracy wit nin I/50-tn of

a second. The response time of the bacKena macnine on small

relations is lominatel by communications overnead. Tte

minimum response time is about one second. S, of tne tests

conducted, tne 1/50-second interval is sufficiently

accurate.

'p8

-~~ ~ ~~~ Z- T Z . --. - -

However, if tne overneal can be reduced, a more rrecise

measuring device is required. Mot mainframe oDeri.inp

systems provite a cioct vita a resolution in microceconas.

This is not available in tne taciena mactine.

C. QUERY SCRIPTS VERSUS PROGRAMS

Two methods exist for performine oencnmarK experiments.

These metnods involve tfe use of query scripts anc programs.

The first of tnese simulates an interactive sessior, acceb-

sing the dataase. The actual terminal input is preparej

aneal of time and stored In a 'run-stream' file, Known as a

query script. The host operatina system Can te instructe

to bbtain its input from a file instead of via tne terminal.

Thus a series of tests can be collected tocetner in a

script. Additionally tne output can be redirectei to a

file, removinz tne overnead in communicating witn a

terminal.

The use of batch procrams involves much more of tte

programmer's time in t e development and debugging of tte

proeram. Development of oatcn procrams aiso represent a

larger drain on tne nost's resources. This factor coul!

severely affect testine at rany installations.

Since queries must te interpreted wr.etner tney core from

a batc job or a script, tne use of tatcn proiramminp till

not offer the advantages of bypassing tne query proc-essor.

Therefore, tnere is some question wnetner or not a tatcn

--........ "

prorram would provide superior performance resultb. Tis

question and tne ease of development of query seriDs

suggest thnat the use of query snripts is tne aesirea ret.oC.

If oaten programming offers a significant performarce ir,-

provement, additional testirg must be perforrel usnln, tatct

jobs. Here it would be wise to run a comrlete oattery of

tests in tne interactive environment, f'llowed ov a subset

of tnese tests in tne oatcn environment. TOis suoset snoul:

be lesinel to test areas wnere tne oatcn process may ,ave

-i its most impact (i.e., tne iata return ti-ne).

D. INTERPRETING ThE DATA

Tne interpretation of data is a very important part of'

tie testing phase. There are two reasons for tnis. First,

conclusions cannot be drawn from raw Iata. Second, Timely

interpretation enables tne persons conductine tne experi-

rents to analyze tne results and identify further testing.

A collection of raw aata is very nard to interpret.

Therefore, any results obtained snould be rrapflel Immediate-

ly. Grapning the results ITmediateiy allows rapid

identification of errors and unexpected results. Peiatea

results snoull also be Prapne toretner. For exarple. ll

S hie results from a query applied to relations of aifferent

tuple lentfn and relation size snould be rrapnel toetner.

Once tne raw data is analyzed, tne graIns may te re-

fined. Tne graDn axes may be varied as appropriate. For

. .. . - ..

example, tne response time may oe grapqed against tne tuple

lengtn, against tne relation size (in tupies or tne aner

of blocts of tne storage Space o7cupiea) and acainst tre

quantity of data returned to tne user.

$

I

m5

A°

VI. CONCLUSIONS

A. RESULTS

Tne results ootained from testing several confi~rratior.s

of a relational database machine nave providel a oasis for

levelopint a general set of tencnmarx tests fir relational

latabase mactines. The bencnmar~inz tests nave been mostly

macnine Independent. Aittougn a testing met-ocology is

provided herein with enoueh results on certain confieura-

tions, ailditional testine is necessary. This testine snouli

be performed on otner DBMSs, preferably witn different cna-

racteristics, to ensure that tne test is complete and not

macnine-specific. Tne results of testing selection and

projection operations are described in LAJ. Eesuits from

performirg tests on join operations are described in [ib.

1. General Resuit

The response time na$ been sn wn t oe proportional

to the time required to access tne data. This, in turn, has

been Shown to be proportional to pnysical size of tte data-

base. Metnods useQ to reduce tne aiount of data to te

brougnt into tne Temory for processing (suca as Incexnre ani

string compression) improve the response time.

The response ti.'e is also proportional to tne arount

of data returned to tne user. In tne case of tne P.RLM 1100,

tne t ime reouired to return tne lata is t me lareest

52

component of tne total response time. If tne necessarv

information is obtained via apgrecate f'unitions, t!ie .e-

sponse time Is greatly improved. It is not Dossitle to

letermine now much of the response time is aue to tne racc-

end macnine ani now mucn is due to tne host. Fowever,

loading tne nost lefinitely leprales tne response time. An

analysis of tne res;onse tire under various ioil 'onditlonb

in thne nost may lead to a iistinction of tne nost response

time vs. tne baclend response time.

The time required to process queries and tne time

required to process data in tne memory are relatively small

for thie RDM ii . This may not te true for other systems.

Therefore, it is imperative tzat tnese areas te carefuly

examinel when adaptian tbe proposei tests to systemrs wln

-ifferer.t arcnitectures.

The results of tne experiments snow that DEbSs do

qave ctaracteristics wnicn may oe measurel. A well-ron-

ceived series of tests can measure an instailation's

performance, and rain an indication of its performance ani

its 'personality.' Tnese tests can te used to compare D3MSs

aciinst each otner. For tne DBM1S implerentor, tnF tests

also provide a netnod of aetermining poorly implemente!

parts or the system.

565

2. Rellrcn -R111

Tne experiments wnicn nave been perfirmea nave

supported two lifferent types of stuay. The first is tne

actual measurement of tne baciend macnine's performance

(albeit, with lignt load and few confieurations). The !DM

1100 provides a comprenensive (altnouga uncomplete) rela-

tional model wfticn successfully offloads rkMS tasirs fror tne

nost. Since evaluation of the macnlne was conductea simul-

taneously witn the resear:.q, the tasK of evaluatine it has

been accomplisned. Some areas tnat nave not teen fully

investirated are due to tne lacr of time. Other areas t-iat

nave not been fully investogated are due to inrorrplete

., irplementation. As an example of these areas, tne use of ALL

in a retrieve s is contigent upon tne number of attritutes.

• At one point, tne use of ALL on a relation with a larae

number of attributes results in only an error rressage. After

installation of tne accelerator, tne use of ALL naits tne

command. After tne accelerator is removed, tne problem of

talting persists. Another deficiency notel has teen tne

inability to perform an inequality join.

B. A RELATIONAL BENCHMARLING METHODOLOGY

Tne proposed set of bencnmarT tests nas four pnases.

The first phase consists of preliminary tests desiened to

identify the best method of measurinp the system's response

time. Tne second pnase involves isolating tne qlfferent

b4:

i* 4

I m m#| m l mm m lmm.. . . . . . . . . . . . . . . . . ..- - ., " - - ,' " -. - ," ' - -.- '. ... "

K7 .7

components of the response time. Tne tfnirl phase investi-

gates the system response in specific areas. Tte fourth

N phases verifies tne results ootained durine the Dnases two%

and tqree.

Most systems nave at least one mecnanism wni-n pro-

viles a time measurement. Initial testing is designet tc

identify the one wlich optimize tne precision ottained ver-

sus the ease of ootaining tnat time. Once the measurerent

metnod nas been cnosen, it is cnecred to ensure tnat It Is

accurate enoufn to provide the necessary precision. It is

also necessary to ensure tnat tne overneaa involved in

retrievine the time loes not reluce tne precision of tte

measurements being taken.

If the necessary precision is not readily availatle,

thn tecni4ues are availatle to increase tne preci ion of

tne results. Tnese tecnniques involve performing an opera-

tion several times and calculatine an averare. The tect-

niques selected must be reviewed for side effects. Tne DEMS

-ray have the capability of Internally optimzing performance.

For example, the order In waicn tne queries are subnitted to

the DBMS may allow tne DBMS cacne memory manage,rent to

reluce disk access.

In thne case of tine RDM 11k0, two 1lirerent metnoas

of measuring time could nave been used. The first metnod Is

to obtain a time stamp from tne no t operating system.

55

. .',

Alttough it may have provided sufficient precision, it .as

not been investigated because of tae otne7 metnoas

available. Tne seroni method is a time stamp ava~iaole from

tne IDM. A built-in function supplies an elapsea time

measurement intervals of one-sixtieta of a se-ond. Tnis

provides sufficient precision for tie Teasurements. Since

the elapsed time ib a sufficient measurement, tne rore

precise measurement nas not teen usea.

2. Pnase Two - Comprnent Isolation

Once an adequate method for measurina time tas neen

verified, it is used to measure tne performance in several

specific areas. These areas are tne four components wnicn

* are involved in all queries: tne time of process tne uuer'

(i.e., parse it), the time to acr-ess the lata in tne aata-

base, the time to process the data in ne memory, anj tae

time to return tne requested. Tnese components may ve

considered tne D3MS'S primitive operations. These primi-

tives lo not tate advantage of any metnods used to improve

the response time of a given query. They merely measure tne

performance of tne nariware and software in perforrine soe-

cific functions. It .tas been statea that a perforrancp

measurement of some aspects of a DSmS is realily a measure-

nent of t.e operatine system. The operatine system does

effect DBMS response. However, in tne case of a nacKaenc

machine, this effect is minimal for some operations. inile

56

,* * ' * . - - ~ *-- . --. -

this issue may be debated, it is not of interest to t.qe

user. Tne user is not interestel in tne reasons wny a

system responds poorly. He is Interested only in tne fact

th at a system performs properly and tne fact tnat tne sy-

stem's performance is better (or worse) than tnat of another

system. He is most interested in thne possibility of ocital-

fnine a q uiccer response time on nis applicatlon.

Tne system primitives are measured by a set Of

queries which isolate different aspects of tne response

time. One set of queries is desiened to return tne same

amount of data from relations witn the same number of tu-

pies, but navia different tuDle sizes. Once a tunle is in

the memory, it takes the same amount of time to proje-t one

attribute from a set of 100-byte tuples as from a set of

2000-byte tuples. The difference in %ne response tirme for

the two queries is due only to tne time necessary to trine

the tuple into the memory. Tne times required to process

the query, to process the data and to return tne data are

tie same.

Tne second set of queries is designea to measure tne

time required to return tfe data to tne user. These queries

return a different amount of data (in bytes) from prcjectior.

operations on tne same number of attributes in te same

format (i.e., strings, etc.) in relations wAlic are of the

57

. . . ..

same pnysical size. Tnese restrictions assure tnt tne

access time Is tn.e same, the processing time is t.ne sare,

and. tne query processing time is tne same.

The third set of queries is aesiened to isolate iata

processing time. In this set, tne queries return the same

amount of data from relations or the same paysi-al size

(i.e., identical storage requirements) but naving a tiffer-

ent number or tuples. This provides a measurement of tt.

processing required relative to the number of tuples pro-

* cessed. The query processinR time, tie tata a-7!,es tirre,

an! the data return time Ere tie same.

The fourth set of queries provides a measurement of

query processing time. For operattions on relations of any

significant size, this Is hard to measure. Even on small

relations, it may not be significant compared to simple

system overhead. This set of queries is more complex tnai

tl e provious sets. The queries are constru,ted to allow tre

effects of tne time elements (i.e., the three Just measured'

to be subtracted from the measurements, leaving only tne

query processine time. Considerin tfte difficulty In ob-

taiing a precise measurement of tne query processir.g time,

it may not be worthwhile to determine this value because or

its small size.

'S, The previous iscussion indicates that the query'

sets are independent. However, witn proper planning tne

query sets may be combined with equivalent results. In tne

eraph shown in Ficure 3, one set of experiments provlies a

measurement of data access times and data return times. Tne

set also isolates tne constant query overneai .14ricn

includes the query processing time).

Figure 3 represents tne response time of two que-

ries. One query selects five percent of tne tuples ani

returns all of tne attribute fields from eacn tuple. Tte

second query is identical ercept that it selects ten por,,ert

of tne tuples. T.e queries are both run against relations

wita 1'-byte tuples. The relations vary in size from bk'V

tuples to 10,0 tuples. Point A on tre graph represents

the five percent selection on 1o,0 tuples. Point E repre-

sents tpn ten percent selection on 50 tupies. Sirce eac.

of taese queries returns 500 tuples, tne time to return tne

data is the same. The overhead associated with eacn ouery,

including query processing time, is t"e same. Therefore,

the difference between tne response times represented ny

Points & and B is the difference is tne access time and tr~e

processine time of the queries. Point A represents a re-

trieve on 10,00 tuples, wnich i! U5O blocs of lisr

storage. Point B represents a retrieve on 59e? tupies, or

Zbq disc blocKs. Assuminp tnat processing time for these

queries is insignificant relative to tne access time, fore,

the difference in tne two response tires is tre time to

access 250 disc olocics.

.559

t %A'..XE.. . . .. . . . . ..

-~ - -- -------- T.02

ua

, I CD

EL t-41 a.

.0 M!W c

Io oI~~~~1 Ad~+~~--.

cz 0--- -r Ce --

600

The overnead for all th.e queries shown on tre erapn

is the same ani is represented oy tae common inter'eit C±'

tne vertical axis. Ir the time represented by ?oitt is

adjustet for te overtead and tne time to access 25V toio-cs,

then tne result is toe time to return 5( le-byte tuples.

Therefore, tne use of one query set nas identified rates for

accessing data (in blocs per second) and returning iata fin

bytes per sezonl)

3. Mae_.e Three - fst em Response

After tne time elements nave been measured, a set of

queries are performed which measure the effect or metnoas

used to Improve tne system response. An example of this is

tne use of indexes. Tneoretically, tne use of indexes

should improve system performance by decreasina tne imount

of lata accessed. However, tne index must be accessed and

processel. Areas of interest here involve deterir.iine at

,nat point, If any, does tne use of indexes become Impor-

tant. Therefore, performance on Indexed relations is

measured over a wide range. '.nat type of Index (i.e., clu-

stered or non-clusterea) provides tne best .erfirmance and

wnat are the trade-offs? 4nat scope of indices (i.e., one

attribute, two, or more ) provides tne test pe-formance?

Tne latter question may be one dependent on the appli-ation.

In testing tne RDM 11VO, it as been note. tnat, if the

Inlex is lefined when tne relation is oelne nreated. tnen

the size of a relation witn a clustered index is larrer tnen

61

a..

r, .. 'w ' - , . " ' -- k . . - , " -- -' :/ , . ; ---- - -- ' < ;-L "- . -

oQ i¢ • " . . . . . . . . . -o

- -. - - -

tne size of the same relation If tne index is defined atter

tne. data qas been entered into tne relation. This is re-

cause tne loading alooritnm assumes a normal distritution o"

key values, wnile the data is in tey sequence. data loalel

has been generated already sorted.

Adaitional testirg snoulCL be performed to ret a

'feel* of tne system. -07 becomine familiar wit tne sy-

stem's capabilities, tne testing personnel snoulI be atle toletermine Interesting lines of experimentaticn. Areas of

interest Include tne overnead associated witn projection

operations, the use of string compression tecr.niques, anl

tne efficiency of join operations (in different available

memory :onfirurstions, wnen available).4. _ual 1ur =-tulli

- The last pnase tares plac-e after t.qe otfner test5s

nave been reviewed and ;rapned. Analysis of tne previous

tests should provide some meanineful results about system

performance In general, ani in particular areas. Tne veri-

fication phase serves to perform tests wnica verif' or

.isprove tne analysis of tne previous tests. It also pro-

viles an opportunity to reao any tests wncn appear errone-

-. ots or suspi'!ious. In tnis phase, adlitional tests ray taie

advantage of the flexibility designed into the syntletit

database.

d' ". . - - L-.- ->.. .. ,,. .. ,.. . . . . . . .., . ,-

u l lk d md j, n '; L n~ m -- 6 ' , N ' :- ::" : : " '" ' " ' " -'"T _' T,, "- "- 'i" -' - ''" "

C. SUMMARY

Investiration of the performance or several conriaur-i-

tions of a bactend relational database macnine nas provitea

considerable insitnt into what may be a sound basis for

general performance testIng on relational DBMSs. In tnis

thesis, a metholoioVoy has been laid out and tne initial

pnases to be taKen in thnat mreznoaology nave oeen lerinel. A

complete framework for suosequent phases ,&as not beep fuily

developed, out their contents nave teen dis?-usse.. Pnille

tre tests described relate to a specific series of relation-

al database macnines, tne basic methodology may apply to

relational database machines.

63

:.. - * - .- - . . . .

- .7w 7 . ' ° . . " . . . . °

APPENrIX A

stem Tables

1. Databases - catalog of latabases in tne system

2. Disics - list of disis Known to system

3. Loci - used by IDM for concurrency control

It. Confirure - information about serial and. parall l In-terfaces, cnectpoint interval

b. DBInstat - information about current activity in treIDM

1. Relation - catalog of all objects (relation, view,

stored command) in tne database

2. kttribute - catalog of eacn attribute of each relation

3. Indices - catalog of inlices tnat exist in tne database

I. Protect - catalo of protection information in the Ia-ta base

5. Query - stored com.1ands and view

5. Crossreference - catalog of lependencies among relations,views and stored commands

7. Transact - transaction lowrine relation

S. Users - mapping of user and group names to user ID

9. Host-Users -mappine from nost ID ant user ID to IrM Ir

10. Blocialloc - catalog of disK blocks

11. DisiUsave - database allocation

12. Batcn - temporary transaction logging relation

13. Descriptions - user lefinable descriptions

64

a~,'-4 o,. - -2 ' '''''"""""""""-"- " -" """ "" -""""""• -"-" . . . .....

APPENDIX BDatabase Generator P-rogram (CMS PascalVS)

MA 1110 4

1GOLU c

*L "Q '44C

OC&Gtalw AloW-fz 49 C~z

9109,l 1- L U

ftw I ..Un ftQ

I. I9 4wO NJ1

llou0 CZ 0 LUS.Z-sLU Ge _

1100 11OLUO 1-. 1.2

UZ-- 5*.W)c _ 40.4JAC6-6LU Wid Lusc

Z0 ZLU z l", X4 ac 1.-LUAtWa0Wu LLILO0 zn O ft Oz

7* 0 z cc Z" t.4LUIs w4 W fto 4%1 f c

0iq Ut~l 0 0-Is I- QU (A IwLuft 44W lLUn 0.cl I. C1-B 1.JalismcSU16 OMW 0 9 WL >j -ft ~4444CLU *'- A CL I ta) LW.

_Z 4vw 0 0. .f 1 >_. Ci (.3La .. Wv Ce 0 ,*. IsmW LU =U

MC-m %sow4 Go I c WUU $me- >.~ ACI:) . 0 C13:3) _j

q4-.JU so acC. I-Cc )m ZU "IL *0 0 4qU*

CU0 N 91.WW X 0 I- N Luce 1. Q e, U(J4~ I'm O W W Z XCI ** 11--4 (4 o M U*"OC

P4. Lulo 04 -4 U(4A* e~d-pmqI..wj Go L" .. uC LUC* -# O - U '.0.0 Zrm '@.40 Go US at UxZ 4 1 . C4 U A

LU)C llWP 11owMl OLU s.44'.4( IJ Ax 11--Z*4 -4C10.m I OJ.* .D-2 4x -qq -CLU 4 n . - - :l Cie

qqgZloZ* o UZO.. 0-.~ *e" 1. U. ZWW.."a4 ILI-)44 14-. ac ) 0~ LukWWsoc =-UZ -4 I-4Z *0 IL9

3coo -- 1-1 ie 0 C3 FmQOCK i-~ p-OZZWuta Cie 42,G OLWL ubow NI~ 'mL~ 21(. I zq( 1 0 s Luom qqi) "IIIWU IU W

4r LUCCW Gq OU44d 4 .04Umm.'IWS WJ IJ Q. Lu" 1JLUi fa I" 3XSMUwel Z.L LU m..Jzw "M~ mlwl"MzAe.41"ll "A = OAM"f..W1." t_ .IL0 L0-1AG-W, CCL I .J .4u3-w f0.W4WW~UJJ.= lW.-iP..J "I 2N Z14,1094)C OM0111 yj qc4~4~w.J:=ol4 OC W.-.340.0.m-ci lu A0D -9 LUO.(n Z4peat 491-- Z.0J- 0IZW ~"xu20- LU UW

~I-OZ'. zzu.j Z4441C LU4490-Ca. ^1143C n" z0-i

11--iieI u1. #A LU Q it 4

0 WO.-lew z . 0cc * Ulf-Ws 0 *

65

''dim

ai I.0 6

P - .U 99I LU

Luu

sI--

I LUq

-M ad Lu

ju zM O0 0- 44U d

I- *so U I.6t

A00 1- 4cu

-ft& I- Q~ cc9 j CLU

I *"aJQ U(. U. Q

LUZ L'lIZO-- CLw wo

'aLU 0-00 Z 4L J 1 .2 0 w I.CK *Z 01 I- NUmc z a>JW. o - L I "ccq f 0 1'- oo 2UO - L 9 - c - l- cM6Z *6 i- z C& i" *L U. iu-- I-

0*t A- Lux 0" at Lc x - e ;1--

GINW% qtp 1 .4 XVJP cc -~uj aP. ks aW -44 X0 0%0.t. . q t. tn L al-

VI W_* luu X.~-0 )LU I.. LUj -7: 6-(.5) 3*-G C~c LUUJW.£6.. * U C (3 qz4 0U JI LL

*L D Os o u acup l- 1 s -1-ea Q. 2U cc"- :

LU z C9 uLU CL c

Oli .4126- 0~~66

~ .. .ZU W 50..Z UOI- LU s2 ~fl ..J -J 4 56-A

!'4

U. X

01 03

-.- 0 CD. Ce .

1-- Iag 4 0Lu!4U 1. -0, cc0

tu ). -r z

2 r Qz2.Uac 0 go*~'

I ~ 4c 'LI L

1 2 WI 44 at " 99INS .UL -f t LU 14-

£3LU e -4 LU -6 J i

42 44

(aw (al -3 JL4 49 ii.r LU 4% U iu - C LU LU '42LU- j 0 Ge 1

49 ac 02 cc 1CCe1 4 cc C. 2 '.Al Im 0.4 F*1 04 ZA 0L

0 * o ac .12 ac cc

CDigZ ac LU Imp . 0 LU

LU U w.4U -w LU so4 '. LU I.LU t.

I- LU I.- "44 _j (3LU LU-ac -Jl +,awa. a0 d-I X 4* N

PO 49 "OLA N LU LU" OO111W. U u* JLA. "-0 31( LU.. g LU V-- '.

lag, I-U . 4. CD *2 LULU 2uod 444 4 B-U aZZ I- 4 " 1.~ w~ 0 g, mm- I-L NIL 4a s 4

(3 U at 4 . '411 NW c~ 0" (4 .4 -0 LU L

9- (2P- 0O CMA"O a" LVI-3 4 U m11. cc4 00 £3 -LUJ Z-Z20oI 4-% 49- Io 0% MA 0q U. I w

2 L IN(W..4'. ZLIJI-M I-3 LU M-4 I 0 Cie M= Wn 2lw U 0,0J4"qLrZ. ZUI .U Ul-i LU :3 ui QU P

W C.0%2WZc. LU L20tJ.J cr. LU AD3 Im I-CL kU 4 FU. ge .6 Pm-rtd. g U V -fm

44 Owi LU me LU 1. 4 .0-401 .409 a.9 V)Z-4wcc

LU CL 419 t. I.- a 2 L

0C 0c"q 0 C2Dc .43ZL

Mm . 0 "A. (ao0 L-t "9a

* .a. 0 0

ILUI~- LU 144

I"t uj

I -J LU 1

I~c I i.

U. cc Q0

cc 0 1- ox

I~r ua15Co1-~.i LU.%

I 4 LU I U U a

4% - I"~zUz 4- -MID.

IL .3 LU 0 I L g ccIom

LU .3 mU. U 440.43

zI Z 004L 4cu0-444 Cie -M x C e Z.Z

Ji ac -a-OCLI- LU 00 .U P"

L b LU 49 mp L U tu S- .4PO'n1aUMac so a. usZ 0 W CL

us I- Qs-o Co4 .- LUWu

sat LU cc LU LAJ **c lof

us41 t- t- w WWWn

LU U. o 4 Z.4aU4 L ' - us-11 *o J -041-t -.

5-4 - .. 0.4 ZZZ 4 0

1-i .CLU 1 5-1 L. U .

LU.. L LU V) t.-2 z LUa I- ac 9il, mL

Q IX t 2VO ll 10 LU Z 44ou .* 0.-- a

C21 1-U I- Iu 0- 8 CC.I3* ~ ~ ~ ~ ~ UL (DU~Z-3.~L

LU - U LU1USf .4U LU m 1jjj jJ-.," -L q

jU Q . .4.21-2 C9 U. Q XWUW W K*-L .-

Us44 39 1 1.3 ZZ~ 1 ~IL0 ~ ~~~ "0 LJ4qI pm C2 atUQLLIJQO'@-~~

0 ac LU 1- cr. LUI w w m *1-14

44 LU g do J LU 1-

.J J ~0 LW Z~ . LWWJWWW 41W68I

-. . . - -

.. '~ I.

tI oi LiL

C1 e a -9 cc

I'm LU 9- 12 4.-C

-C4 (3 . 44 -Lu (

0" cc I aA

-M PO fA "00 (-

g us Ci 9Ca m&q1 81 = 4

:,UU z tj a 0

-be O"Ve go CD

'U4 acw t A u

4. zz - A . * 3. a-0OI 61Im Lu 0.9 CL wa UL 4a

a1 1 420 A 0V - n PD20 u Lu

a- a-I- Oew cc m pf 2O LD-c C34*cI-CC (D 4 .0 ac 0 4 cc W c1 1-. w z c o IN z a. t OCCiU

aI 4A Cie aJ a-c ILIL isa w Ou .1 a31. dcc0 C Lul aii

AMI aa qtt IA A3 Pw ADI0 " qq c c.m d mIs

I~u o" 0%0 LU' pa- 09-wo Pc 0 um o MW

*cnu z- 03 a- 44C'46Wi (ax~ W LiL u"L UUCav I- Nd~d

49 u i l - 0 -lt lu 9 -l 04 .4.'"0 jia JL 44cLLi - eC C 4 Luu

"M a ot wa1- .u q LU " =LUL

a 0m 04 CD 0 .(3~a L0S aca 2u z 2

UA bO (L.. LUa-

69s0 A -g,. UL

... 4

a 00

X -9 Z L

IIL LUI- LU 0

4AL w go-

IsOU '-4u z =u

%I- oU n -LU

-JQ ui PLUx " LI .4 0I- -.44 t- Q0LO~ca W. US- 14 P

0I 13 XX goo sk LaIS40"w cc" 0A MW X

WOM= MVIQ.IJO 1"Wa 4c~~0 i n ~ L 4 ~ 4 4 0 0~ in-adZ~ 00~~IAac z-L~'4 1 J. 00

Q tJW IJUUJIj NLU NN 00acac~ "01or-"44In"NWinW ~ c3(5WLU -J. 0

-~s -Meg4 4ZZ' zz -q= QU LUE ( JUrd

a" LU .JqS I- 4

0.0.04ui ;CI- ula jw zo.c=.c-M .. w.. Lxw oQU - - L

.4 az w z up Wo ZQ O OUCIIJ ~ ~ ~ ~ ~ u I~ zII IIg IffZ 9j LL U"J -..MZVW-j a jo Patp qq c A) 06W LAD W UI W

LLU XUi s. ca U U ASU

-P w W SU LUso Pgo "os O I LU o- 1.. 4L W 4p aC-sdI "41 9-0 WJ - -W UW L W L

ILUWW DI ~v

& --- UWU-.O LU~ z X 0>' Lu.a. . - 00~ U

I cI L LUL

LU CL

700

b4

S 6

us LU I-I. - I, *C3 Lu

0" b" Lu48- - .D cc

f - Z - LUL 4 ZIm 4 $- Za 0 ZW I-

vs co V)c 0 DIU LU

U. Umn.~81.04

0 e ccZ

0 0 -onW 4LAS LUO U6 o t114

LU 0" tu zacW I--a U) LUa

L=Io LU ="Lu:)10dcccZ LU u).a

.1zL wWLU4

.1L e 0 Z LU6

Q .1- 0 LUAl LUPWV)Z Z :Z) Z

000-0 Z C 0 Pt IZ LU 49 0U)**

% PaU LU _jZRIZ 3- C9 Z C.1-. 40 4 sluW

ItLU t 0U LU3 CQ > Lu cct

w Q9 31 A0 a WcaeLu ago* 0 LU* df u .&NLUfa C20 tu;Z ZLJw 0 LU.J- LUZGC

.&ACL Z aZQ ca3 =wow X ImP-0= -at LU ?-00. 9""dft LU .4 A moo" I-ALi 4. I.- -

-j Oft a :)r J= = I- I- z L4W 049

LU -4 1- tu~ so Ixt- 0 0 zWu 1A""-% LU U,

1MU-1 0 4 UUO-44 If 4 e LU w I-a LU wo- wLUX-o 09- J -CN049 ova~.. ac j N,384 up ILb 4403 -Ia LUs t1-0

NU. W-el so el- "-.- u U LUo so-

-- *= -J I" MM--- w - -j in I.9ac LU-- LU 0M

w)PJ Ze U 4;4Qu U) VAC XX 9 a Z0*OU 4..J UL.J4.4.4LU"Omc LU us 4 4U I 4U LUw-i ddmul LULU LU M LV

ag.ZA *e= LULLISO~LUS 040 ZWt -

*JC dZ WA aace z L- OW j .Q a-CZ71

X 14LIa A

0 0LJ -1J

(A ~ 0 II=I

A 0 is usSQ V) a 0-J

0 xW 4U

cc usc2 (AII

us Ill Im- a WUac 0 4 (4 LU.

V . Li t- 16 J w

0U 49 QUI. o- ac"Z .in W- I A 0" 42 0 -*in m MM $_ CL 5 "q cc

A M, (a , LW 0 0 4CL dim 0 a- fACL:

0 1- I-w SA M I3~ .- U

C19 a *** 0.0.9L0U 0 t SCf

CL. 0r LiI ev L0e-j ~wU 4

4. u" 0 4 W WL. C0. 1. 09 0

0 U6 4 W 0Ua WZ~jxIU 2 UMUJ*

zZ %j U 0.49 1-4.Jp 0 4L e I- )I--7 1 0a w. x- ca 0 owZU. a a-. wQV qu P 0.t 1-0 W . *.MO.W.U 4c s" 0

0..M 4 C M* O-Vegl-T 00 w, I- SLLA 2. 0.00 Zo a 0 .'w w 0 2 60.D0V -

ZO 4M mg 09 cc )w0 I- 040CL 1b LULV @& w 4 t 0460f o a-nj~a- SM 41 $W2.# A.&~ ~ I 0 1L4" g..4Iutil.." 0. Xe.W-a& q

POM ip 41 .. u k- C6-9 JO130 l 00- Mi-. %4 -A. 49%944fctc I H It &u;; *.P 00 L a 0 0 Mot

1110 Ew z. 1- ._. 2. so '~a- - *LI IV.'-w I o

U&40= 6MUA(Dw s &U U ww e4mAC AX0 CC a-L> JZ. .U -P-_v LU1 0_. ZPu 0£ -Ul j i s- 9C . a-.$ .a 4 .1

IJAN us I"U zM WJ- I-- I- MP- 1-1-e UK Z SM -0 u

CL PMA SO a- c P". 0" 00. .1"02U 000 tD ICC 4q XI. cc cc cc GWCCU i

tie LU CL CIDU-.S

72

.0:~ S =0 -

I- utI Ux "4

Igm SC 0 0U

"0 - -ftWM

"I O3 I-1-L

I-i (AlpLU

443. LU L

3U 41 0- _-a- 4i WS S. I US JI-09 I

-. 3 0L 1. ql*. 01L0 Ire * 1-w I-

1- LLU x S S 3iA3 u LIjc (I. LUC 03LCL0 340

I0 LU-S ftU LU (go of-~qi ac '40 I9 Ig0p a I-US 0 :ZI- t1

909 .-C ULU ~UI~Of LU LU 4S

*~~~~a 0 0' UU WeC3 I JI-4 01 goo U co e

a 4 W NO two 0. QW. ac t*ON OLDc tu LU0 I-IN 0. I-fl

a1 A L4*4 L I* P- Z- 0 0 "goF-4ft 0' 4U0 rtsW ccI-c - u . 00.. x

1-02 89 0 4UW 03 Om-VI 01- "OX A1-4n W LU .0"440 A. wo 0 0 No 0 -0 a

" 0 *40119 II- ;c-"cc.L30 I. IZ

soNJ Z9 aI 0-- 44 ) - 0 .4 Q aigOw - US 0W6 J . LA. A 4* 4c -0.3uzop-O 0 C2 a o LUM904 4U 0 U LU Qc LO

I"04 fw40" Q U64 I" I"-4 L. 04mw

aI & L *"ucte **a C ain ue~ i

LUL .1-OW0.4II" CC Go 09

30 (9UQ.' LU UA 4 41 LUCO N

LULU L.aU zwp LU

I-- U J a- CL

Me u US3 m AJ

C.L

.9CL1

73

- ~ ~ ZVY -So' 7~.'7 c-J 77 . ~- .. -n - W----- 71.71 -.

.4-4

~c- iLUN

LU QC Lu

00

0~ L (uj

LU 'A0it

qq u I-OS. ag

LU VU

CC 0"0 a*0

ui ~ LUI- 44Uso a~ ac 4 (,LUa r 'K 02 '. ANO 0-0 LU go oot * .'KW44 Pm L 0 L

-tga'a at LI 4 CCS.

5. us VP 'K 'i ,

AM 33%LWW 3 *-"a.4 1-! '.low 4a Put1 0U.4L4OO0aem~ 1*44 'KZK N A- I r (D .Look jf4W ZV L ; 0 soj %Ku LU ft*.*'K~g I-s ZL3(J 0 5 -'K'Ke j'U s-UL goILUO

"40-K .4 LUZ CU'UOPL zt' "Uq Us s- cjgz5M6LL ImI WW4nri U

accx o Ku'g 04es. '-QO.a____oz z **-f Op. W4 Ckt 0 I- I 4t

ac 0. 0

C9 LU z C9 ~ ~Ot ft-d CL 9U 4 JU ' 4CLL.. .d ~

.4 4 pQS I.J I.J. * JN'~'. J..74.

L0I. Z U 0 ~ ~ .( c U '

'4i- LU 4 Q.W~mI4 2W ~ L

IN 7

'33

I U,

I I-v

I 0 LU

toma "0 Q.4 INDU .63c

ac I. 2 I.

5.4 ~locc *A A~

Q *j CA 0 4

* 03C A zNJo oWaW) -Z: at 4 '4a t -

800003 02

ZZ. - 0 I "Plz Un Po0

z.. 'oa tQ-i m4 0 1" 4ALU~ at aj 40. .29L Lom LO& s

u0a~ Jac2 4-7 zz , ?W x~

Z ~ ~ ~ ~ ~ ~ .4 N f L U aOA LU 4QCL o-N0 N 24l-3 14b (23ouLCPZjZ 4

-i g*Zie 0~ 4SC.o Wl- LU Q4 O54004~ z 94X f 09 ZZ .4W LA

002220 00a o.-. Is C U4 Lx9. s 2 4 0 b 0WS c

cr. ~ ~ ~ ~ ~ ~ -10 2a 1:4N" I( 0m~r f 3 v$4" 2 I Z oeLN MA 22 2 bl~ 4K - t 0o. I-*-ZZ"a4Z i

9 OL4 di Io-oac Z 1019 Ac x . .. W;p(3 ( W

.je..Jiu 24 Poo LU aCC -5Z 10- OWZOf~ w **"J0. 2 to -0 i -- 4- -I I LI-tZ.ug. 33" &4Ga Uj* t4 4q % t a.= 0 s-axo. " $.- l

z No.2u~3B 0.1 'No ac ac IU z C 01 IIM 1 10- W4

ta N UWUSA 'a. I v 4% N-A) 0,-D £

204 z u LU Is >. Us -4 0£ '.

cc LU 0

75

A - -

4LAS

.40C)

0 0 02 .J Iac

-G to C; LU.

(3 1- cc IsIPM £ 0 b- .oL

:4 1 00 O-.0

FU 4 1- 1

Lu A Lu LW Z9

as sea.. x. at * .J-

4U I.J t4wo .4 0440 QCLu SZ~~~'~. U *I II

a. vp Iu- I ** - L160 Lu £

s04~- cb-N a c( aI a 0.1 a. *'I-f.J *. ag: d I ~ '3sso Pd

s~cacccu CLXCD. 0. OZ L z uA I- J.

I-4GC *HU 2W ZU IA 4A A ac ;a 4O-2Zc tLuWo Lu :3sg-dw L

~ ..- Z4016 pa4z~

* ~ ~ o-~ Li WUI X. A M ~* .2S ~ g ~ I~ Lu & s o A nZ%1 - C4 ' -'su m - S M C

Lu I . 4L UA 4n Sta 0 OC AO.

S 44 ILU AJ L L&I O ZW 21t 0 0" 1-.J.Jp4 LU _Uc &"ug x JW AIQ

L .I ZL&. -i u W44~4 s.v) j V LU Z *

,a O£ * Lu 4

76

44Z .7

a .

-. 4 44

-. 4 L1

LUM~S eezz1. LZ0A C.e== 0OC c

000 z 4OZ OJ ZU 0.% -u

4/. 8. Ij A + UU44

O* -4. 490 W 4U ur q

Z. pg*J4 A a Jw0 !-J 49IU

*~zmM I i ~ cC

a- 4Aj-3JL -~-~w p 1*12Oo 1 4 U

.1O43 09 5-u ocda N 4.4L p~cg * - c C S. N $

I I SM

op- M 1 4

tu vp CX qCX 4W~~'x =L (

2U -- o *'.0 * Z.JZ - U (- LGL~ P-11- "a L a " m p-w -*I

0~*. vs "I-tv *W~wzw *V W W2 usQ- SM 0(0 a2 zUS UJa2r.'.

CL93 CO UA cc LU 6 in ~ 40 u I

~' q ~ SM.e ~q,,u- ~ g 4w .~ W 4 Cie

4 -.Z ~W. t~I 2 I'..JI~ ~~g77

ALA.

Poo 0

LU pow cc -A

LU - 4

pq 7 LU a- If

fA '-9 .0 4 on (

-o c -444

in $40 .1(L -M

of -N 1 0.4~ 96 e

zL NU 4.)JII LU 2'-.owC4 JL 8.4 0 Z * 41C c

loo L.D z -U(As n a t (A -. 1- .1 "Ol4'. u

aCC "q -J442 0 J-ac

4smi ac wg-0Q AI.-9 ac 0 -In -s P a- ac P" .JoeL us 4 LU a- **U..Z

0 aw-"M, s ow Z W9P4W U6 cr 42 ajN ~i CLU Miss 1f -,%Anq cc cc

w. CJ0 A4 a in 4 " so li 0 "u

qugps see Is-U of, a- "O) 5oo 4 rZ I W a 0 f a-4 o q .2 * mu J=.. k. tj ad o"w a Q. ~

alt. tal LU .q 0 j zoa - Is~w 1- 0 js

-4.'v .-4 s V# oe 4- Woui N4 Cade sq U a5-Av La

JgZ- 0 1 4 cc 64s ".-W2AW W aL.Z4 W -IUIP-4-0WWW us 414 SOW. .JVW 2 4'-4 1-00

I-vp @-cc (a~ I- W) itLp. Ga LU W ul L U IIWa 4 £-a z few N U Z L 0 COMM

Z a I.. A.u P- La~ . 39-1.-I- LU 14 W- 4JW zz ZWJU. LU 0. LIuuJLfj (at 'i 0J 09% z~no. p09.lp9 ** Q. 4 "" ccc2 2 0f 000 (5t

m)* z LU 2z cc J 2jLL LUs twC3 a. sV to LU

78

4V

- - -- - o .7 7

I.- -

0

CLU0

(.5 C

atI LUj

?-cc 0 0

(a .0 49 OZ 10 0U I -ZZ up .3bp 49 - Q ca

"~- 0q(9Oz 90..

.4g 21L #-~ 0.

cc 1- 9- UU N -0 c .Mif 4 w 4. CC-- 'U U* C o c Ccc

4'. 1^1 440 - C3Z - 0 qN

0% "- -i L L )- P 0 q 0C*go L

0PI ac.I0. LUZ- 0- 0~ 0q a .fw o- (01-u 40 1" - LUI cc -u

P60 WJda0 6.24 I I- so u UaUPP1- I 0 NO P4 P" R 2'-. MOZ

XC 0 U*1- - I- (A 0 0*0 'Iaa 0 Q#

WUI..W 1.0 L2 0942 ZO UZQ.P~4 U z U) ~ q 0 gA. O I.. g

UU t- WI LULa0c 0. 02

"=OILU 0 ww Q 0 * :U ULOU'2' 0 0 0 4 'U -

*W.I LU W.U*.~~~~~U 0.~4- - U ~ W Z Z44 N I

.10 4 . U U) L() n.* n. e o. L01-fe-. IW4 - zq79

viMv~aJ 9 % la E~* '~&'I& '~

z7 T

LU WU Lu-j W

9-'Z 22 I

1- cc"aUa

Poo 0" P"

.4(3 r L

-- am I I4

" 4 I-. Q

wo LU . 0 (4 .9 (4 0

.44 W4 £e 4 #A '

- Z I w pI I- *L

cr W' -J" W4 r..j'.0-4 41 9 4

2W L4a. WO.Zooft C.go#

-. ~ 14 (It . 41 4

"" U( '. W N WN C4 W-40P

Z (3 "d. 1 2- go. z4 z I-4 90 O .4

Liiui im.Q.c LU CCdj CFi~ . .

*g" W 4$"- o(3 93 P

LL do-o. %j Cc~.( LU - ~ C I.ULU " w " " f~ c.

z IU.0 a 0 I~ I' z"Ch. 2i~.

w wz* W. . W* x *0u

w.b,

80

LU

'I-

LU-5

1.J4 .. z .

24frU U. POP .

zg 0 " &U3

I O k 4

L -MO - f aN4s

j LI. Au

p i - I MbW .4 U . -NR -

ul s. 1 t2*0-

0J 6"; 1.

j ., t A"O LU 1" 1~ 9".

4f~o s

PO** I. Og- .4 M in 0

3c-J4 s0.q u 4 U 0. 0 .9 j (20 LU

ac> 4.. G. uz .03 .4.ap U.~ I. U .ZP tzw. t4 "4 p"afl * 4L

U.~ 'm ~ IL Ne- .A Q.Nou '4Zj "'4'.5G 12

-.j LU I4 Wek 0 4 WO 0(.D 0

2. se4 LU% P-o .-.U=. 0 ac... "" *I .us ~.ista-ee-44L~ )ft3. s ee*-c 0 ~n 4 &.L Z .4 -u'~

U"CL -4 ta*.O .4 LU 1-n

LU I.4.4a* *U~~~~~U IbU0b1 ~ LU U'4J. t .'Vta

- *40 . 144fl ZN-3 W . ~ tub Oiamr '4U. * 0 *.0

MCI& Z * wtI& '. kj" **.. d&Ii..bq _

64~ -q1O4d4 U t 1 Lu.6 -a,4 LU 4 0f ft.1 2.4J1-.~~~L cc1'.1'4 - flU. .141 ..JLU1-'.Q-' usI L JZ

*~~~~3j Q~fiD.~U J-j cc '4IncDSU Ab ~

LUW.4V 31 LU .4~S .J UU 4% Q£41. d4q

640.~~~Wa u~ (Jac - 4 3 4'4I

iUW~ ~~~ Cf.J .44LW2*£3. 3I3Q UIZ :,~ a-U~u.W4ILU3 IA

8 1

to -LJz L

(.5 X(5 4 0

(n em-Z .4 Z4logoJ (* 04X -40

q* a4 6C 1-Q I- I

a sft U Uj -4 U6 Of dV0. CC0. 9U Q6 AC I aL

ov U*in of- -0( -u LL~

-- & SuSM £0lm 0U C4 b"M 0'.4IW c*0*f4 (#30.5 X 4~ Ul- P- I-" -t 0 1 %

.. up 4c ~ 04 I.- It- 4u d- 0 - _Z -0.- .Z-04 I IU a ' .cAq~ "q x

U, b"L M . UJU Ii I.-. 1 . * I- LI Ja I& IM*- (2 1- ; z-6 V IJ ( q

00-"J I- 046 L- a 0-Z aS wk.J4W'4 49 C% *Ont- 1- 490AL0 Q 4 -" U tr

X-4--i o ZZ O4 l L Z 94 1J Z 04404 49N ZU49glm41 I Itt 'l 9 *.P "Io t 4 C LU d= 4- 0 "Ju I-- "- U

AM.U-=-- - sa ccI 4 00 0 - Ji LU ACL 0 . AI040 W es"0I P41- 1- 40 L 04 a t- "44242. 1-*C. 044N b

ob b s wt O f awo"§.e0. 0.n .Zw-i 4 w 0~ (3. 41" U jtoLZZZZZZZtZcU 90.0.'LU LU "Q "4 % ". wSMUaWZLU IZ

uIjQLJWOW "-30 ca Q 4L -4%4 ". - (a .j.J~jt ca SM..'.4.J de

act.a- -zz wc j 149 0- 4cc LD 00QLLI1I1--IO~ 0.o ZI- .JP-WU .J ..JW 0.0 1WL 2 04

-aLU~Z V,4 441 'W4"%. SM. %,%0 &LU. 1-" ow FA5 LU CLe1 go ell O. N XLUO.L 0Q-

t-,%JI4JJI t41- 1t- "q 4W L- %w 02 1*"'-U) '"4 LL taqWWltllmv(ivlw~ltiIr. I-" Ow mw 0 ?- r" .9CSUW4WWUAW P-M P-Cr CC 0.JS I t- V"C0

LJf.JLcal3UJX~ 49XUJS U)'.4c LU SM .. 4 11 W LUV)w~ 4. LUL- x IA **cclu

. -. aL I5 N1 Z44L NZ 10MwLzxxzxz~.L . LU $O.wtWQ 0e9 "4 aC9***".5. 0.

1-.-J~J~d 1-0 0tt- w (2DPo-LUIULWLUW~o-o ZtlUI-.'.LU Z'.4Lu z~-I--1-11-1 Cg cc4U~ LU Cc LU

82

A .. . . .

-j

I U.9(3 0~z

Z* 2U "u-ftO4

cqw-4

LUU 1 (0W6XLZ ~z

00 odt W 43 " m "o A U c f C0 Wac

Z9L 4- XN LU I

o"O 6JJ I e-. 44

co 0 U -20 -1

eg (.3LU"-a ol

"I- 1- 0*

WI. we *ZZZw A

12-0"~~~ t.Je. I,0Ig1 C31z =9o JJC4cCv

U~~~~d%'-, CUW N-& (.3(.03(.3CW L x o-I .a -. I-18 114jw ca 0. IW"Us.4j

*U 0 .U S Ztp (3 00Z.(3.

- .I--o'0.--O wz p'uw"N*a aj ** 14 idu.I0

(s *. z~ Luu U43 .0 *-'-l zq0"* w 0w JLU o"(2 ILW4 g LaW

0 - 0- OC4 MCC2G.I.-3 4

eOcc -0-1-0(3X-21 g-s-H0 ~ CD a (3 ~ Ip.-~Wew z - u4wwI UUJ.S

q U .. Cc uz CIDwa"~e-.

4 Lu3w. o~l ZN U.J 83

4 *4 - 2*.S. IUO'.-L~-

0 LU0Lu I-

'U

U u LU CC-

4b 4CL (d

0~U -0. Q .

sin o M 0 =Z0wZ :3 Za US 14d 04 Q'cc

Lu3 1 12 1 4"c ZO 0 0. U) Z.WI-e Jc . Lu

.4L J La

Nm :2e *O u a.J= mos* O b= o L .fta I. I. J a 1

I*6 UJW -mo 0a a44geLu 30-z zZ Lu Q SAWI..J= 4

)N-- c 1- 0- ZWU I-@ OM=C rIt-Q %CC LU0 P-. $"C 4%jWA vp i

52 - 1i 4 4 imW M IU60 "9. o #*1-0 0.4WU Lu Gal- Luz fI4c" a 3O mt

P wl -gZ U I' a-jdc f

V)W-C -mJ 4 0. 1 WO -q JIOUAO WI4W .U 0 on 4D 0 1- . "O-1-w I-1 UWO. "4z _j .2 41LL1-at LU 4

* d0 UA.. Lu. *- SOI 0 -4 00 aeg*ZO C .. 0*W404 '11-ZO .- M * J.On 4 OM 1'-i

us- 4q.W 3-Ze I.s-2.WW0-

X;MV4AU1 0-4 "-4 xxv " "m Lu 0 LUu 0- ac IN CC"q aZ2.- 1.00" fgatPs - 00 CD9

4 W-W pJJ0 ..J* I isa wc~wa 0.W14QI-OUO.W ww m .4 U. I-0 C1 U.

*49-"JWO.JoOC3 0-0- LU P-0- a.- t- Z

wg pZ022 2c.x1-4 9ULU I- pULad~ j II*-m""S' WI' w 9 -tu I- aci0.

wWW-1ZD 039 9j 04 '-eZ q 1-)0-

1.- -40- of--5 * ~ IUC- 110tLjr- ~ ~ ~2 -qr$v t d ~A

. . . OLUMCC WinC LUZ- LuLJd vp P-_j Cf

Lufab" Lu *..JI- Luca-. Z

zoc0

0

.484

- .~~~~~ *~ .t . .. ~. . . . .*7~--- -- . -U w --- - U-

APENI C

DaaaeGnrtrPoga MSPsa:S

APPENDIX

C-,j

0

I.- .C 4Cac -.1 .3

:1U 0..03 oi

224 101.-g. 0 4g. 0

.6 wU %a AftN '.

ac~~L LI- 0Ltu .3 1 a2 ui 3

ui~U LLIU I

z C19 ).. c qouLA4Ic.4 Z

0 LUJ2CC Ulg a

4MULL Ma t- CM LU - 0 0.cca f '3 -cow 44w 1-.4 3-1

I- C6 - U9 .. C.0 ZI- 9C'-0 z A)m am O..Z 4 1-

*i II 4 .3

1-0 *O j .oC6Lz- 44. .5%I'- wO .36

I-: jui c~-u WSz 0~n~ W~ z~ "N -

16qc N.. pWI,'4d F5-A cc LU J -IU zi& I" cc 9.

I X N ZJZIW N 4 m6-ZC...J.II -44 $.-'43.W~ I-S z06 - N uI (J IHJ II0.W W wlJ-I I 411 zuA -4

pq COWAW I 05ka"0.-w w 0Z.l- EZJWua tI- rig gue

SjZZLUi. Z444 W44 0qcwlI-N4u. CL.o0 ow

cc0 09O.L

85

0. z

I., LU z'

tI w

LU LU 2LU LULU-

0A 0 z. *0

LU LU L

*0 4 -:411 > 0

:aD LU N"Ip Ill 00 at B

us p" s LU 06

w 44 1U LU(.z. . C

ac 0A LA Z~~ I-. 4 4l- I-

4 z - em m 140uc'ml -. -cc C~-3

x U LUw No 4U"lN U L

v Woo w4 4 r- 44 1 LL1 w .

ul 4c t4 age *u LUo 9 A*, 20 1 .

10 1- pq LU -40 4w p-.usL LL u LU LP 1i 0.I u

Zg C- CD I O I u. (3 acI'm W c.cma .4 g g U imLI .. ' *

(a U0u Zc 0u I- tu Ac u -son I, u

~~ LU tu0 s I * QZ'

wo IA =us -t I'm LU ZA A S LU LU 0uita eM Q Pot *& t. I" QI W* e .0 aM LU L

CD 0. a Z a. 4 0 - 0 Z3 cl

LU U. z o LU Z LU L U LU aceI- LjX c 9u z c LUw ~ ~ ~ ~ ~ ~ 0 (3 0. (3 C Lca0wC

I- (.58L

7' - . - -

3L

*1U

.0.

00

O1. .'U, C9al 49 1

vi 0u :

40 00 LU

40 ;a211u I.-

a e . Cm a 5 2

m 44 Luca I- U L u0 - c~m=-Xc(D e

t ~ -40 0.4wl-Q Q L

44 -AL ag soe Q A (64I at544. C6 49 Po cg2-sJq" W A

I t" dc m W." AM "1 w-0 %v P4" zl 0C&

g.. Lu usj t 2 Lu oC -(vc c 1

5 C 40u 04 $. Lus INj L-em " Lit -4 .05 Lu " m Lu I- 4 IJ'-e jtLu

0- 3 £41 "0 t-0 eml- *.LZ0(3l-z 411lzj

Zf a za. OLee Ia 4 L~ W(0" 0 W4C44Z041"W I" us 39 c". 600 0- so UP I ILUU

al £LI A Noe 46(2 u IO I I dc I I 43 &L~%u -4

0~ (5( WaO LH ac0a0IaI c (5 at 49 £ z. Lu )P- JW,IA a.~ cc~ L4 L c usCL I.

* ~' ~: 87

C-~~~~~w d7 -4.. , d -

-A4

ujz 3

(a a

S . 0 9

U. z

I-

o 0c Zoo*LI

- 0 09 2

010

Iwo 01 a t

C10 0 LU Gf - g 0

01 W ..( I.- ac ftS.zD

is .. '0 01 aO-o.ZS. * LU .1 ==cc I- I -j tu

LJZO.I Ig 02:)4- x O

X. 01CicN;U. -4in0UAW- cc 01WWN Z Z-J- =CC

a J@"4x 0 4 "-4 II 0..011P" 14 g Zao~ 0- X ~ 0"gz S- NiZ N 4

IN-1- 0 -Us 0 A-J m1 ogo:) 0=00ow -I- qi*m 0 of.-5:2 CL -ftIeeJ o01 3C4*. * I-

Cp I 01 .0 ...-09 %NZZ A Q 24l Wag N 0 0 . - OZ N CZCIo 0

"al We.2 .fl4V a J .W 0- 1 2 009 0

e-J6 (J- SJf #2J.*- to " 01-.m~0f-mf1 zxWj WIG CGILa0 =3 I 3'-...i O(1 Z 01c Na -0 qq0q'410za 20.0.0..4g"I.- wz I0.WWONO-W I" cc N *q2IN W (3C

ULJJCJc um **Z I- .0 P-LUW01c 0 1-00-M LU Wo 1 04 S" c

a, 0~ > tIg 0.zL 0.

88

- - .7 7 ---

ILI(IL

~z

(Cm

I. G* 5CL LUu

as"Izz cINzyto c0

acIZ:

IO 0"$U LU L L

:;*I Zoo -Jog 02 0 0.z

1-' 0. LU .U I ac.

.sU z~ z s- . .

z3 -cOC "q 4 acN-q 2 0. 0..j.- IU LUU " I I

*000c LU .owt 0 of J *0- gs10 1

-- Ia-CL 94-ziz LOZ .AJ.. LUx

z wU IZ-P .. LU l16ima. LU Z~ LLOC" Z W ill- C WZ LPJ lzl i.- u

a %.A %0WOU 4u pC LUA W N S66o VSm4. 'w -

cf z LULU .- C oal- LUW.42.00 C3 00 UW 0,4 ((.5 0(. 0Z

.1 a JC6 1-~ P-L ZUJ JX 4 zJCLW jCgxz ac W-039 ZZ"13ol" I LU

00 . OU ObIZW 5.1- N"mZL L J

UA~~~ VwIU3 L

0*

' 89

E-A

2e a

us 0 ac

0041-J- xnJi -3

zuw-iWa (.9 00 S z 0

us uss 0 1

I..I- Zz $ 9 0cc mm a.Nt u O

Ch. 3 CL s ".'., 1- - 39W-

00 vw ICA- W .3 ae"*U

1 400 2DC 1.I 4*. 'i ig .. .( 60 acF

ac Z %w .I 'No.22 o~t Io atupzsjj 0 I0 ~ ~ ~ ~ ~ ~ ~ .~ U Im t 4 nt -tca Uu"e.9 34 SCCOSM - on*" 100 31a-l 0N mucac 09

cm *"x j0az 390mulms

t IL '4u E4U 4 was ZqCO-. .Z~ujw so 02W L~j&U

C3 CM ge o - sw (a 0;c AIL WO 450 4 W 1V I" ac49- ac w 2U

423 a.Z'4Si4 U j ca4 w m

1153( 0'.80.~ U 02 JU023(8iZIZZ@90.

~ OW I. I '-0 *muo EI

::cr.SUu

.$:i7.

PO -0 c

I0

- U

* LU uU

Us .§J z aUsL usI - " .

cc _j 44 Uswe

A Q LUt- ui-% I

oU 44 jin Q 4CCo 1" cc uZ> Q

ad 0_-t LU I

_811-131 LU f -- Wa gs oz 0 aso +-L 49 .Iis I4e0" JC-U

0-6 U. 49ua3 N. 44s e 0, j 14 4

40 A Izp '.4 so 1 0 twat U3ip

Imz eXLI. LU *. ".*4 440 44 Oa -a42f-3 -mw 5.Z 1- -

0 vW 04" is" %0. *O 4 .a4 L" i % ow- rum4-LU ggZLowUjc44 ( A04 I.*i x~w Z 4 ut

Jwo LU (0 -- *CU, >2 L IVZI- _a(_,%Utz3c"O KI-s A1 & ra I* 44a 4. (I-o t-. 4 #

mW tiss P- 44Q PC "O4JW6P" eU

#-* 05 OUU'O p( 4aZ .42 O 2~a VI p.4mc z. ..t N .

CLO ~ ~ ~ ~ ~ J 944 40*"M 4 U 44 e

*~~ ~ ccD 'Msd cc.., * ~ ~ a. I- z 4 11" U .4 2z LUI 4 .- 4 q t- U ct X=1 4 2 -4

*~~~~~ 0'I2'. 20 0 Z Z~ 0 (2 IWZ4

~~*;a M -w zU tISI" - ICL . U . Lis .- cowc

A'- .91

I INDUU

I '.4

oft

0.0-l '-6

I- Jo 0

0 W1

(jas.-Q

00Z "O1-i c k

P~~a1- a0c j A0 ai

00- V) -*- -C

ag MI I P Z

I. 2olo ..J I- M ~ Q2 I Avi IAZ 1- zi - oI -Jo 3S.1 1 0 N 4 z3 0 1.- 2

ac. - t0 P"002 = U-1,1

at- IJ p. I-.a=UJQ Z 0 CI CL z

"4L4J~D~ SO- .4cz NI~f~ IW.4 f

go 0 go W4 tZ 0 U.- 02Wam Q '4U 2I c.0 : N1 1 .00 C sUCt

ae 4~ "moU~C 10. .J LU LUCDM .4

1- .1694 *O *SovzvwseUSO I- 1 S-W acNzcU

0.5.- (i I I cr--44 I Io*'Ul*4j I #- '" N c 1*0 N ZL

0 OAZ 5.-55.15 4 J;A ZLU."" (a 0I LU foJ**-Of 44u

1~ -00 -p Uw=OO5-'4 20 S cc 330

ki ac w~epUa a~ 1-0 ZO LUZO

ZWW aWcOIJLJU xw53~u LU CL vQ O.4Macco ""ZOIL Cacopq cc* . Q LU

00 0 i 0 0 0 c

u Ch co L c LU 0

~ 92

"-I-

'4 4 uj-4 .4 4

I Q0 U.%

IU ft, 14

31 0.IQ 0ci Z M c4PO O C4 494 w 40M- '4C

I- LA Z sob4 V.* c.Obt0 0"4 0Alw. C2 0 U 0j

UA -Jac cc'CL j 0 e i.4 MCA POP460 "". O* A 0- 4 4 VOV-5 lo

Z0"bwI(a C 1# 440 c 0"I PO- ~-00 LU

44 QW UU,0 L COQ z-c

IUJ IOo LU x 32 LU (32IU ca0- I'-LU CI~~~qa -~ so 004e J4 ' ' ~ I'~CF W c .A 2

LUQ U

31 ). 0 '4 .ZI .- e 2 - IN ~ 933

*1W

W t-

W, W

W Lu I-

49 49KJ asUL'-2. cc

to- _ I.- ae LLI 1

84"u W. Lacu I >Cf. I.-- 4

1644 1o 0 .J 4cOv q rUz Izj cqs 1-iOj CCJQccg -OW Lu. UW Lu 0- (

ac. wee. P4jQ c - P444 W-1

'IMP :3 CI-IACO Q4tI- -JC - I .Co

NOI cc ac -6 W U"L W co to~--oC 44IAI If ju. a 4l"OU6 0 *0 L .OI6

"4 5 Wq u I..ag Lu _~u 4..D ZOaz0rJoeacw 31aoe 9A .- o a. W WLLu C I- **Catn . 1 I Lu

Q Po 49--CCLCA If4 W4. 4 0-.I ZU. 4. *II.0 *J Iq I.- - j

I- .*4 -o

PoI'cm Z.J *a "*LUZCLu C@.J Qu - tIDqqzrr :2w( (2u02ra 4o4u -. -0W

z WC. z~O W VI3U WO 4L4(..L..W.s -1 We cc. W.4~~* i- -Jqt Zf W4wof~5 of CAWW~5 m -~ ~ -. ' -.J-Zm. OW *'4JW.J2- .0 ..JUj a.J .. f-- ieZLI ~~~ ~ ~ ~ ~ .L~ '40 * .50 u* *'9J0 L ~'wg

U.1419~~~~wja 11j > .4L44L U'15NJU*~ ~~~~~ ccO ~L *1l WW ~~

I~~~~~~~ CLU WV~4Y & ~ ~ ~ *J

%!..u 3W ~.J0pd ~.u 4 P I~.I~0tfU94L

f 4-J O ... J S.wJ ElJI) .Ju-q.4 * - .z

HD-Ai33 651 DESIGN OF RELATIONAL DATABASE BEPJCHMRRKS(U) NAVAL 2/2POSTGRADUATE SCHOOL MONTEREY CA V C STONE JUN 83

UNCLASSIFIED F/G 9/2 N

11111. 5 J '.-4-

MICROCOPY RESOLUTION TEST CHARTNATIONAL BUREAU OF STANDAROS-1963-A

,4.

S", ... , ., . . ..... . . . . . , . .- .. , .. , . .... .. .. .. .. , . . . . - a ... . .C C 0 --. ''J-- ' l m +m h,,,,,,' , ' ,,

-. .b ' ', -''

100

:1LU 1-

LL -. IM

I o-

_jd a a

too ( 0 co a.

aca0ag 0- -

§fl -OC(

I~~I 'S ~ ~ 5

ZLL D a+

*&K -490 Mu.4~ NCO~ t-*54

Ja 4.. IOe WA j e U.Ci

0.64J3109.0 4 S .9

F,4LOS .dfw OS P" f&".4 ,v w '

k-.I 4"llf I1"U a0 *Z -. eeA0 .j a U z0:

I~t 3 0 C eLOU...oi I- *I- meti IISM'00". w.. 0 -4 -0 .Z . .

.. M g-Z-4 L

ma C19 ac 0im4U~% j 09 dc 4 0 WI-

wod&IS M US

9~(D too.4 4 1 M&,e~-.uI I.4.

LU z 3ui z WccI . c OILCe~' MU4 tI +w.0

'I 0~ Ses41I g4I~lI.~.00 ~95

.40

AU~Iso

dollLU

C6 I-

0 4lI -~ ~ L

S I I L

so *U 0 U o"

0 4* 0 C6 0 4

""o 41-0

cc.0 *Z z ( L-st0 t0 '

oso- 5M Lg LU *C-= acm C.4C.2 -c G (" j t .1 t 2 't N )

49 "*.loma Z 4 LU -> n um F

.1. ( -*4- AU.' 1- o = ftc* 4 Ac c-os 0 w -1 *

(3scnot Mgoez a O0 'A Z 10- 0-.wU~f 50o U* m- Mul LU tca ~ Z. NWWWsiz -so "Z. 1- I'>

-a-19 go 'i4cag .*M...aa Pwi= V Wn -2 3w j

0 _o~ _ s ZC.LWIUxt ZN (5C(S- ac- IN 2tw -le- Is,*0cz -(5aase m2 t41PMl. 40 LU I (5W ILU 0.1- 310c:

COHWLQ=UQU94ww U

r% ILasso 1-tp- 4 0 1 1 PAP" 4 U Is, '

*j I I I 4LU-.- 0 ~ 1-.'.."dn0. *.2 LU-..0I 59 fi .9 1w- N"a2 ~ ~ ~ ~ ~ ~ ~ 1 Z "W44121-*Uw4U ~-I ~~

0o 0u-s ow a~.e I Zdftcc L W' - 0.1 ~ ~ 4 0 Qs""0400,4149"wW ' a O

Q JZM CZ AK Pq Ot.Z*VA ~ dZZq LUQ LU Q~c

44 w -IlQ.W-W U - 9gUU&'6 4 4% 15 * 0Z ~ ~ ~ ~ ~ ~ ~ llao Ia-U 11)gUA1- MU0 IP

Pa WZ 0 O 1- 9. 40J0.U4'U Z 030ma

.41-0 1-Eqj

39 LU 60004~-J-4

0WOwW-4&I4J ~ 5Z5LWW L 50I -- 196a

W. I4.t U W W W u L i t L u t j L

N . N N N N NIo f* N " 0 4 p . 0 0 " "44 i 4 nf 4 l14 o cI9 u 3 1 9 3 9 9 c lcai - 1 - 4 - - 4 4 4 -"RIa c a c at a a o c Ia~f : l : P : r : : r : t z P : P

W-0 06 N 0 0 0 lb 0 0 0bi 0 0b 0a0 l

*JCW 5115 WON E IX P U WO ux tNt IA LU *j UL u

IC4- T0= UMJ!"5 U"19U. UjQi WJM UJOj jU, Jj j~j U j U:10 aco C3 ama macOoaa am am a aaCWAA ~. Ia A. aJ M 4A 1 20s .M42 0A

Uwe N"0P 04P N " N" N N"a00 N 0 N N N C N q

ca ;341.J~. -t -410 34 ;11 .0j -g -t g -o g .40. go q.J wUf n wu f 4#1 a Ul Wi W~Ub tU d-UU I Wj~tW-Vpfi A j af Wf WP-J Uj- WA~ 0 ~ h lo w = m M m SeUJ W Sew NeW )IW )low SeWPU SeWu SeWL eW SeW 3

~aS S M. .b M am ZL ZL Z4. M M l b dA.M . 2 41-ft"UzU zU "z " " .3 P" .35 KU aU ZN RU 33 XU,

C02 0 a;04C1 2 zc. 4c. J:W:U:93 a:a:C

QQW Ofw WW 1jW Nwft La. Ca- o- o- ca- "- U Q- L3- u30Lr.%& 0 0 t WU,

I 'L3

S 0 Mae 0. 0, 0u. Q (34 C3 0~4C' (a i (Ji 0 J. (Ji (J 0.40

* 10~C ~ ~ S w L ~ Q 0 ~ J J 3 ~ 0 .

6 ~ 3J C C C C C C C C C C 97

-~~C K- K .... I

In A- In

a C C t" to *

* 4 N ON N U 4 N U7all a. am4 .4 4b 11 07 ob4o

a-0 mj m.4-0-

.J .1 j -a -J -. a .j -jU&s ws u us aL wo ak w us~ Uki

Iu om Iw I-uJ I-QU IW O-W I.- owt t-W O-Wad~ wa wa wa wageu wo usW% wg wN

S J . . 4. " . - -1 .- J 4.4 _" i~ -J.4t W~~ ea we wb We lbUS4 Wl W. ft M e WeL 0M

cc' 04 ow amam am O 003 916 00 00

M LW 9NW &W LW 46W MaW IW IL% NW

P" _C "* 4 e 0b 04 0"e Qe "O e Qe No0 q

w4 041 O _ 141 U41) 141 141 I 44 41

5. J4 Q4 1 4 -4 4.4 W-A I. -A4 -a 0.4

%w U40 U.w LA-.. L3a% LJw C3%0 Qwo Lam Q3w *

0 acN~LI~LI6~qutd~wumdum4IJm44NIt 0 cc

4 P" w4

98

................................................* .- ,

SI'.'

S

*1.~ IQ

.A

II

--.5

S

ISS

;~ IISI

'I~I ~

I~:: ;

000 0

U,oU ~ 0 VI SM SM SMo 0 0 I- 0* * 4 - I- s.q- Z SM LU'04U. 00

~s*i ~

.2 0

~*: Iii 99

1;''4

V *.

.LA

OLOaO'O

S. u

us 3

10

A . - -A.. S

#1:1-IA.,& I

'4 1I

-'I'I-'I

.' II.

4. I

--I

AlI

-'I'~, I-A.:

--I;2 j

A,

-A,

4

A'

ILA' US Z US U,

'UII. 0 0 W 0

tie - adnu~ou. oaau, 0w~mnw 000

* )'G UUNtN).&U~@4JNNZZ

I

4*

101

V% % V V *~d*. ~ -'~~ A..* ~~** *~ ~ ..~ ~ ~,

S APPENDIX D

Valuaset Generator Program (MVS PascalVS)

'ac

is US0II-

LU ac

LUus so

0"~~S 4J4 a q a

OL6~ Z -p C 04

-4 .sI 0

1 44 U w (5 o ( -C4 a 0&D

:1 z IbC a-u uz 2 a U LU004 1U) w 9 W w 5.M -hl 1W

440. Au to Ow cc -w - a pI

V4 ~ ~ ~ ~ ~ C ;A4a-,C I s . '0=,

U3 20s. 0 dc AZ Lus3 04 9auz. .s . s "0 WIZ M c

mmU 1.- 44 W £ %a $.- aSO4J 46%0 %a& I"~ vo UL ' 4 & Z

4,1 344 NU "ft zi us zsUL . usa

C44 oat LU I6 1- .2 is-- P-L usu U -A-w1wo Q6 " .J4 W U.' Wee 1 u ~

usU 0. 0w se Ao JZ - Zs0~-44 .- c c us W - a 44C. ZU #A3. L c tu u

_J0 CCw.~@ MU -. ae

84J 93C mm M3JU W6 mmz C30 3" .Mm£ IL cc ail cc 44 3 cc. aw

vsx~ Q ..o dcW oZIAM. %

LM44 Wu us ziW Q 0.t3 U c Li ZZZW*

JWIJ 3L 0£ 4) 0£

102

Okla'*.. **..... . .

.~U.

IL u

- I -

4wI z OI

LU LU W -=qts t

z" -j A "0em Z C00 jw 0" "2 I-

2L aU 0- ti£wt05QC 4 a..J LU

P=9 9 . 2 - 2~a wo.JJ0*. dit q LU P-" o*cLj 4

b"W In6UI O4Z em 6d LU-LU 0 -oeowk. JUA " =u P2 4W U I-

IND lo Co S-.JW 1. *I*S tn&'4 I

96-" dimu ju p.- l- I-o zliu w.4.o Lu I Iw

LU .dW"I *. W.J l W-L J£ zL 39acN-O" (n.

.8 46-J *. ,11 w114 .2"W Wi 14W 4(a U I LU5R6430 OlIN 092W 4mwk~sco q.)9 3- W o -ZLW

S- U go~SWS0 LU~ C- LU cc-U'~b~

ZWJW .JZ.J2Z~ ~~13 WA .JU W

WwJ.Wa WWr ~ a~ w .W ~ 11

Ii Ci

UL

jg1 64

p" a.-

du If~a* do*

UWU

*~u 0~ 0

*u L4o W r

ac 040 Z Z N I3#6

P- 44 actimmul 44H 96W %.444JM w a 044 UI~ OVICOO0\ I Z 6.4

Iu w

Ub do lumt

Z4 104

ii APPENDIX E:1Valueset Generator Program (CMS PASCA.VS)W .4

I ~ U0 M OU LU U

..W4 6 - Z

1.t 4 z IM a L* L£ 0 z zW

1W xaiz00

4O 0Wt a W a

w.1 wU 1.

I~~W 1 4 t. u.C

LUO Q 4 L..j. C

Q LU-4 LU .b~*2 "0- C a~ 0O UJ coo ::N cc

X-l t- a -UA, ac S W UFA4

U I~Ui.I Wq~.t~0~

444 agW Uo.... -.'e .VWeIIdjC J z

2 UJ LU(. a L 2 W 424.0 0

21- N 'W 9 W 20U 9! 1 z 00hJ.00j4IUZ

clu M Z. aU 1U ' "q

VIM a. au4sP XL. u"9 $.. 4-W-W - 3-14-pL

Ul" VZ A-4t. J'- p"1-401041-1-

aI-lu >a 49* LWe 1251 0 > usx14 Mi" -LU 403.LIg .0. P" .Z 1W

us U6 B 'a L1U' Lj W L44 L 51LU W .JOC

WM~ 0 LU6A .0 ALZU -j I-W. -U4- I I-. 4505xv z.a Oq 29 U. 0c j 0.4 x. . q

W4 ZL LWZWo

IA ,*, UM*. w u L6 n. - -0 1- t- o- I-*" ic CL. *i Q -o f 4 P qC

zz C& so 4% WA cza ac........... a...........* * . CC

.........................................................................

Jal

.3i ccSU -jIL

-. 5U(AI

t- uJ

*4 I Uda Us I.-

I- . At ac*U zo LU Z 0 -

X. U. -U Cf U0 acL Uc (0 L

cc LU tu 01- x -Ci CI0Ut L l Elft

5Z LU xU I- *CUO

*i da cc4 LU-Lu atL O z LU cla. Ui.Cg z L

dite D #--n 4dt U I-Iic - %m Ix 4 u I-n j

Xv3 agU LU d.O4 X ud LU *

20 Z4-4 U. No x I". :24m a Iia LU*m NU *'O I- CC Z lU 2t u

I ElA LUaf.41 I .LULU l~ac ~ ~ ~ ~ ~ ~ L ZOt tw£LU(.I-5g-WW-~ u m zz -OI

.4 A2N 44%a W A id act '4n W.Z Z i

4A & u*LW-I fC MW ft-l .- CIm 5i 4~ 1-0 M X Q4- LUA X

w43 I-C Q LL> - 3ca. U CL LU. LU *.

WLJ UA at ~ LU z c cu2 e.~E44 C (m o. CL ccU2WL

LUEl I- - .E10 WJ6

- -- L o V

'Ii

ij *0

3.L

U uJ

".39

.4 U)

-j I" x

0. II df

ILU Ot A i4 Q Uhus"

CC us_u OLI.04 LUt-N J .

lIQ (= I- z j "0 0 .JI- 1-2 4c3L "00- _0 t

Z 2 LU -* to -ac -- " z2iILU A. I t- a. * rt2_c " 1 .

0-j 0 o OJU 0 x " 0 a

0I "t. *60 .. 0 2 i L I pq -Q'to -ft ac cc *DJ- so ULU _j Q0I. i

* - L 0 - @000 p- .J .01 V) (.,

*l- 'Z '4W sI U. _, "o 10 LU 6"

-27e . 2 KJ UA 4%W -20 X lW W 0 A-W.0W LU 4 -O2CCLU6 - .. 4 4 2 1-Jo -

4w J; aU LU-Iw a mc14 t T9*t 0 -0

j WU Ge m ' (6_L 1- @0 o m U) GU) -f 2.-Vr~fe P rO6 %PO.W0Wjff UL0 SI 4 W-J& tJip 14 s-w0

Wo- miss L - M Us ^U I.- ac2 itJ U. I-LU Jto 2 U- 4 Qx2w NOW4 c C - mU U ix .-O lu *L "40(21

to@wt0 LU.JWM 3 QLLZW C4L 1-1.4 W4% L -40U L U lIa U a 1 4 P.IJUZ& ZG G 39 1 1.-1 -0 at N~ a CLMO 4

f 4JJ to PO Cuillo aj W1 U54. I- *ILeI UI- Ld-jOxaw tu zl Is-i xr actW Usi up* to -it- f 1Wt Id3 0 Ui@

*j0 W 1I-'.1-W 4N 0 UA CCU" t LU WZ ' P-hA ..JWLJ NC WIND ac4 JZ iu ac 3ZI I- p1U4 U)a (ca- U&M 1

CU &S604IW= 13 =q.JU. '41- C'4 UI cp" .A0U 6W I aI-"

zWw U) U4 21 ae1 L.11 z~NZU4UI-o i I a '4-UI- CL WIl -IU e O4)..

2 L.JUJ .JZJZ~ W U) ) -. JI-11-07

.~ S

* mlt Ci

IIwsz *j

up LU

ClLU .4 - O

z LU LiLU aZz A

I- ac 0

4a l &- L-

-j -AC'CI 0- L UI

z .4z LUULUCK

c~ 44 -j co"O X

Q6 2or L w go -I..06W G0 4=Z OWoz LU c~j Z

Ia 4 ucX2rcc LU we W

ue -.0 LU "W ZI ZU wIWIp A X4 weU IeU P- (in..L- ,

P4IM4 ZU JLUCDO"

I2 cc 0"42 EZ we .. in-im'4 ag 4J' 4Zin'4WR A2 *U&

44 I.. z~z 0 0-1 Z w ZW Zae

'4 zw wU. to ~ LU cc ~ IAP w.~

~ 2 1-1-108

List Of References

1. aOL Raniouts. Amperif' Corporation, Chiatswortn. Cal.tornia.

2. Britton Lee, 11t. jofLwr2 Eletelence anual V 1.3, September, 1981.

3. Ryder, C. J., EeIncra1i1 2 DataASe1 Macn1nes'211-1111 111 SpRIZ-2 !-A Database kamjnistTa tor

~ on~jtj~,M.S. Ttesis, Navil Fostgraiuate Scnioo1, 19b3.

t. Pordanowinz, R. A., BO _.qjt26KjnX 1A 11.

tionai Database MacnRIe, M.S. Tnesis ,Naval Postegra(1u-ate Scfloolo 19E.

5. Crociter, M. D.. jf~~ma:j an R2i Oeaonnae-

AI22uLL PA ~ 11 '1111g .S.Tnesis ,Naval Postwra-duate Sctooi, 1983.

109

bibliograpay

I m,_.rtl. s jguagli. q.ulef, April., 19K.

IBMP LcaalS_ mrorimlng galg, April, 1981.,

Naval Postgraduate Scnool Tecnnical MPomorandum, Uir?E_ M1k IS IL N , Joanne Eorart, March, 1982.

Naval Postgraduate School Teennical Note No. "VS--i, User'sGuide to MVS at NpS, W. R. Cnurcn Computer Center,October, jiFZ.-

Naval Postgraduate School Tecnnical Report No. VM-01,Ger.s guimde to V M/CMS a NPs, W. R. Cnurcn Computer

Center, March, 1982.

Ol

4,

. °

,a , .;, -/ . , ' .- . . . -.- .-. - . -.- . -. -. . - - . . - , .. .

INITIAL DISTRIBUTION LIST

Io. ccpi s

1. Defense Tecnnical Information CenterCameron StationAlexandria, Virrinia 2?314

2. Library, Code 0142 2Naval Postpraduate SchoolMonterey, California 93940

3. Department Chairman, Cole 52Department of Computer ScienceNaval Postgraduate SchoolMonterey, CA 43940

5. Dr. David K. Hsaio, Coae 52Department of Computer ScienceNaval Postgraduate SchoolMonterey, CA 9394

7. Paula R. Strawser, Cole 52Department of Computer ScienceNaval Posteraduate ScnoolMonterey, CA .394e

9. LCDR Vincent C. Stone, USN 11229 Sam Lions TrailMartinsville, TA 24112

9. LCDR Curtis J. Ryder, USN, Code 52Department of Computer ScienceNaval Postgraduate SchoolMonterey, CA 93940

10. LT Robert A. Bogdanowicz, IjSm, Code 52 1Department of Computer ScienceNaval Postgraduate School

"onterey, CA 93940

12. LT Micnael D. Croccer, USN, Cole 2Department of Computer ScienceNaval Posteraduate SchtoolMonterey, CA 9394e

13. Commanding OfficerATT4: Doris Mleczzo (DPSC West Code 034)Naval Air StationPoint MIuru, CA 93042

I.~

13. CommanderNaval Security Group CommandkTTN: CDR T. M.-PIJTo-Si, USN (Code G350D)3801 Nebraska Avenue, N.V.

*i Wasnington, DC ?939e

14. LT Linda Vilmaler, USN* 3026 Bromley Court

VoodOridge, VA 22192

15. Curricular Officer, Cole 37Computer TecnnologyNaval Postgraduate SchoolMonterey, CA 9394w

-. 1

5,

-1..

FILMED

1;