A Platform Independent Tool for Evaluating Performance of Computing Equipment for a Computer...

39
A Platform Independent Tool for Evaluating Performance of Computing Equipment for a Computer Laboratory Takashi Yamanoue Computing and Communications Center Kagoshima University SIGUCCS 2004 Fall Conference, Baltimore 12Oct.2004

Transcript of A Platform Independent Tool for Evaluating Performance of Computing Equipment for a Computer...

A Platform Independent Tool for Evaluating Performance of Computing Equipment for a Computer Laboratory

Takashi YamanoueComputing and Communications

Center

Kagoshima University

ACM SIGUCCS 2004 Fall Conference, Baltimore 12Oct.2004

Contents

Introduction DSR, a Benchmark Test Tool for Distributed

Systems• Structure of DSR

• Node system

• Applications

EXPERIMENTS RELATED WORKS CONCLUDING REMARKS

Introduction Purchasing equipment for a computer

laboratory is not easy.– Requirement specification, Designing– Money

Hard work, time consuming

Equipment is usually a distributed system… makes designing it more difficult.– Hard to see the total performance– Simultaneous operation …. Heavy traffic is on the network …. Heavy load is on the server – We want to measure them

Distributed System Recorder/Player (DSR)

2. DSR, a Benchmark Test Tool for Distributed Systems

Records/Replays Real Operations Measures the Performance Replay the Same Operations on Different

computing equipment Let Every computer Doing the Same thing

Simultaneously Platform Independent (Java)

2.1. Structure of DSR

group

teacher ' s node syst em

st udent' s node sys tem

group mamanger

st udent' s node sys tem

st udent' s node sys tem

st udent' s node sys tem

st udent' s node sys tem

st udent' s node sys tem

2.1.1 Teacher’s Node System

Corresponding to the Teacher Records the Teacher's operation Replays the operation and Records the performance

data– on his/her computer terminal.

Can Direct the every student’s node system to do the same operation as the teacher's, simultaneously.– by multicasting the commands which are corresponding to

the teacher’s operation.

2.1.2 Student’s Node System

Corresponding to a Student Records the student's operation, Replays the operation, Records the

performance data– on his/her computer terminal.

Interprets and executes the commands– sent from the teacher's node system.

2.1.3 Group Manager

At all nodes, the same command can be executed simultaneously.

Needs a Reliable Multicast Binary Tree form of Node connection by TCP

… A kind of P2P Reliable + O(log N) (switching network) The Group Manager keeps the form of the connection

when a node took part in the group, or left the group

2.2 Node System

Main Controller Event Recorder/ Player

Web browser

Programming Environment

Draw

Network

Applications

Node System

Writer's Assistant

Text Editor

Command Tranceiver

2.2.1 Main Controller

2.2.2 Applications and Commands

Event Recorder/ Player

Network

controlled button controlled button

Command Tranceiver

clickenterexit...

clickenterexit...

move to (x,y)drag to (x,y)click at (x,y)...

Application

Controlled Parts

controlled area

2.2.3 Event Recorder/Player

Commands with the time

Messages (show a performance of the distributed sys.)in CSV format… easy to analyze them by a spreadsheet.

2.2.4 State of Node Systems

Teacher’s node system can control the Operation State of Student’s node systems.

State of Local Operation, SLO– Independent operation

State of Common Operation, SCO– Same operation by all nodes simultaneously.

2.3 Applications

DSR is equipped with applications by it self. Compare computer laboratories’ Equipment using the

Same Applications. Applications

– Draw

– Text Editor

– Programming environment of a simple programming language

– Web browser

– Writer’s assistant

Draw, Programming Environment

Web browser, Writer’s assistant

3. EXPERIMENT

Compares the performance of six kinds of distributed system by doing the same operations using DSR.

File Server, Client CPU: PentiumII,266Mhz OS: Windows98 JDK: JDK 1.1.7a(Symantec)

File Server, Client: CPU: PentiumIII,800Mhz OS: Windows98 JDK: JDK1.1.7a(Symantec)

Client CPU: PentiumII,266Mhz OS: Windows98 JDK: JDK 1.1.7a(Symantec)

node system

node system node

system

PC-1(1node)

PC-2(2node)

100MBps Hub

File Server(SPECsfs97:7612) CPU: Alpha OS: UNIX

Client CPU: Celeron 400Mhz OS: Linux 2.2.14 JDK: Blackdown Java- Linux port of JDK 1.1.7

node system

node system

node system

Linux-1(1node)Linux-2(2node)Linux-10(10node)Linux-75(75node)

1G-100Mbps Hub

1. Record operations on a computer using DSR.

2. Play the recorded operations on the 6 kinds of distributed system.

Four of them have more than one client computer. In the case of these multiple client environment,

the operations are executed simultaneously on these client computers.

We couldn’t find out any delay of the operation by the sight.

The recorded operations

1. Read a 7KB GIF picture from one place, in to the draw of each user.

7kB, gif, 1-read

2. Read a 16KB JPG picture from the one place, in to the draw of each user.

16KB, jpg, 1-read

3. Read a 37KB JPG picture from one place, in to the draw of each user.

37KB, jpg, 1-read

4. Read a 3KB text which represents a figure from one place, in to the draw of each user.

5. Write the 3KB text to each user file, from the draw.

6. Read the 3KB text from the each user files, in to the draw of each user.

3KB, 1-read, n-write, n-read

7. Read a 10KB text which represent a figure from one place, in to the draw of each user.

8. Write the 10KB text to each user file, from the draw of each user.

9. Read the 10KB text from the each user file, in to the draw of each user.

10KB, 1-read, n-write, n-read

Results (a part, sec.)PC-1 PC-2 Linux

-1

Linux

-2

Linux

-10

Linux

-75

1 7KB gif max 4.1 6.2 0 1 5.3 5.9

Read min 4.1 0.4 0 0.3 0 0

ave 4.1 3.3 0 0.7 3.7 5.1

2 16KB jpg max 11.6 10.2 11.8 11.4 9.1 9.5

Read min 11.6 0.7 11.8 9.2 0.7 0.1

ave 11.6 5.5 11.8 10.3 8.1 8

DSR Could compare Distributes Systems PC-1’s performance was worse than PC-2.

– mainly depends on the performance of the hard drive or the file server?.

Linux-75 is not so worse than Linux-10.– Cache memory? of the file server machine reduces

the access time to the hard drive. Latency of mouse moving: almost no.

4. Related Works LoadRunner

– Emulates thousands of real life user loads on small number of computers.

– DSR: Reproduces the real operations on the real system DBS

– measures the performance of entire TCP functions

– DSR: Measures the total system performance of a distributed system.

5. Concluding Remarks DSR, a Distributed System Recorder/Player Records/Plays user operations on a

distributed system It also can let every computer doing the

same thing simultaneously. DSR can compare distributed systems

– because it is a platform independent. http://yama-linux.cc.kagoshima-u.ac.jp/~dsr

Drawbacks– Can’t use Other Applications

Future works– Spreadsheets, Mail, …– Interaction with Web CGI, etc.

Acknowledgements

Staf f and Students of Information Science Center, Kyushu Institute of Technology

Staff of Computers and Communications Center, Kagoshima University

Grant-in-Aid of the Ministry of Education, Science and Culture of Japan(C)(2)(09680401)