Future and Emerging Technologies (FET)

Future and Emerging Technologies (FET)

The roots of innovationThe roots of innovationThe roots of innovationThe roots of innovation

Proactive initiative on:

Global Computing (GC)

Proactive initiative on:

Global Computing (GC)

DBGlobe IST-2001-32645

3rd Meeting

Athens, November 29, 2002

UoI Presentation

Directories :: Resource Location

Data Delivery

Outline

Summaries for Resource Discovery

Maintain summaries (e.g., Bloom filters) to assist the search for a service (resource)

Directories for XML metadata and appropriate summaries

Resource Discovery

Resource Discovery

Motivation: (DBGlobe) Large Scale and Dynamic Environment

How to locate a resource

System Model:

Sites that store hierarchical descriptions of services (in XML) or XML documents

Path queries

Limitations (so far):

We consider only XML-Trees (no cycles)

No value queries Joint work with Georgia Koloniari

Resource Discovery

<xml><device> <printer> <color></color> <postscript></postscript> </printer> <camera> <digital></digital> </camera></device>

device

printer

color postscript digital

camera

An example XML-description and the corresponding XML-tree

Path queries

From the root: //device/printer

Partial: camera/digital

*

Overall Approach: maintain Bloom-based indexes to check whether a document (item) exists at a site (peer)

Resource Discovery

Bloom-Filters

Allocate a vector v of m bits, initially all set to 0

Choose k independent hash functions, h1, h2, … , hk, each with range {1,…, m}.

For each element a A, set the bits at positions h1(a), h2(a), . . . , hk(a) to 1.

(A particular bit might be set to 1 multiple times)

Given a query for b, check the bits at positions h1(b), h2(b), . . . , hk(b).

If any is 0, then certainly b is not in the set A.

Otherwise we assume that b is in the set (“false positive”).

test if an element b exists in a set A = {a1, a2,…, an} of n elements (keys)

1

1

1

1

Element a

h1(a) = P1h2(a) = P2h3(a) = P3h4(a) = P4 m bits

Bit Vector v

Breadth (or level) BloomsResource Discovery

The Breadth Bloom Filter (BBF) for an XML tree T with j levels:set of Bloom filters {BBF0, BBF1, BBF2, … BBFi}, i ≤ j

One Bloom filter, denoted BBFi, for each level i of the tree. BBFi: the labels (attributes) of all nodes at level i. BBF0: all attributes that appear in any node of the XML tree T.

device

printer

color postscript digital

camera

{device, printer, camera, color, postscript, digital}

{device}

{printer, camera}

{color, postscript, digital}

BBF0

BBF1

BBF2

BBF3

The BBFi s are not of the same size We may skip levels

Depth (or Path) BloomsResource Discovery

The Depth Bloom Filter (DBF) for an XML tree T with j levels:set of Bloom filters {DBF0, DBF1, DBF2, … DBFi-1}, i ≤ j

One Bloom filter, denoted DBFi, for each path of length i (with i+1 nodes) of the tree. DBFi: the labels (attributes) of all paths of length i. DBF0: all attributes that appear in any node of the XML tree T.

device

printer

color postscript digital

camera

{device, printer, camera, color, postscript, digital}

{device/printer, device/camera, printer/color, printer/postscript, camera/digital}

{device/printer/color, device/printer/postscript, device/camera/digital

DBF0

DBF1

DBF2

Special symbol for “root” paths

Resource Discovery

Preliminary performance results

• Both outperform (in terms of false positives) a same size simple bloom

• Depth (path) very sensitive on the number of levels

• Depth (path) need more space

• Updates are handled efficiently (just the corresponding vectors)

Distribution

Each site:

local-filter: a bloom filter for local resources

one or more summary -filter

summary-filter: merge of the bloom filters of a set X of other sites

Resource Discovery

Horizons

(keep information for up to horizon = d neighbors (as in routing indexes)

A merged-filter for each path: merge of blooms for all sites on the path up to length equal to the horizon

Resource Discovery

Merged of nodes 1, 2

1 2

3 4

5

Merged of nodes 3, 4

6

7

8

9

Merged of nodes 6, 7, 80

Hierarchical

Resource Discovery

1 2 3

root peers

Leaf sites : local filter

Internal sites : summaries for all nodes in its subtree

Root sites : summaries for other root sites

Resource Discovery

Future work

• Evaluate distribution strategies

• Other ways of summarizing data (related work on selectivity estimation)

• See how this

can be related to ontologies (meaningful path queries)

whether/how it can be integrated with querying

Directories :: Resource Location

Data Delivery

Outline

• A survey on different modes to transmit data: Push/pull Continuous (periodic) /a-periodic Multicast/unicast Directed diffusion (communication only with neighbor nodes)

For the 1st deliverable on the topic

Data Delivery

• The different data delivery modes in DBGlobe

Tradeoffs of using one over the other (e.g., in registering services, directory (location updates)

To be extended for D10 (Data Delivery and Querying)

For the 1st deliverable on the topic

Data Delivery

Data Delivery Modes and Coherence

Data Delivery

Focus: How to achieve temporal (currency) and Semantic (transaction-based) Coherency of Data under different modes of data delivery

The Data Broadcast Model

Client

Server Broadcast Channel

• The server broadcasts data from a database to a large number of clients

• push mode + no direct communication with the server

• Data updates at the server

• Periodic updates for the values on the channel

Data Delivery

Efficient way to disseminate information to large client populations with similar interests Physical support in wireless networks (satellite, cellular) Alternative way of transmitting information for data intensive applications (e.g., web)

Multiple Versions: Not just one value per item, but k such values [Pitoura&Chrysanthis, IEEE TC 2003]

Temporal and Semantic Coherency (Theory and Protocols) [Pitoura,Chrysanthis&Ramamritham, ICDT03]

Data Delivery

Clients must read consistent and current data without contacting the server directly

Currency

(x, u) RS(R) CI(x, R)

Currency Interval of an item x in RS(R) - CI(x, R) - is [cb, ce) where cb is the time instance when the value was stored in the database, ce is the time insatnce of the next change of this value in the database

, say [cb, ce) overlapping- equal to ce

-

RS(R) is a subset an actual database state at the server

older value OV_Currency(R) = ce- , where ce is the smallest among

the right limits of CI(x, R)

Data Delivery

Currency Interval for a set (readset)

Two properties:

Temporal spread (discrepancies among database states)

Temporal Lag (how old with regards some point in time (e.g., T_commit)

Protocols and their properties

Timestamps (versioning)

Invalidation Reports

Propagation

Data Delivery

Consistency

Degrees of Consistency

C0

C1 RS(R) DS

C2 R serializable with the set of server transactions that read values read (directly or indirectly) by R

C3 R serializable with the all server transactions C4 R serializable with the all server transactions and the serial izability order of the server transactions that R observes is consistent with the commit order of transactions at the server

Data Delivery

Protocols and their properties

Data Delivery

Relation to temporal coherency

Based on broadcasting the serialization graph of the server (or parts of it)

Future Work

Multiple servers model

Applications in sensor networks

Data Delivery

DBGlobe IST-2001-32645