Parallel and Distributed IR

1
Parallel and Distributed IR

• Eric Brown

2
Parallel Computing

• SISDsingle instruction stream, single data
  stream.
• SIMDsingle instruction stream, multiple data
  stream.
• MISDmultiple instruction stream, single data
  stream.
• MIMDmultiple instruction stream, multiple data
  stream.

3
Performance Measures
S
Running time of best available sequential
algorithm ----------------------------------------
----------------------- Running time of parallel
algorithm
1 f (1-f)/N
Slt
1 f
lt
S N
?
4
Parallel IR

• Introduction
• Develop new retrieval strategies that directly
  lend themselves to parallel implementation.
• Adapt existing, well studied information
  retrieval algorithms to parallel processing.

5
MIMD Architecture
6
MIMD Architecture

• Inverted Files
• Logical Document Partitioning
• Essentially the same basic underlying inverted
  file index as in the original sequential
  algorithm.
• Physical Document Partitioning
• Each subcollection has its own inverted file and
  the search processes shard nothing during query
  evaluation.

7
MIMD Architecture

• Logical document partitioning requires less
  communication than physical document partitioning
  with similar parallelization, and so is likely to
  provide better overall performance.
• Physical document partitioning, on the other
  hand, offers more flexibility and conversion of
  an existing IR system into a parallel IR system
  is simpler using physical document partition.

8
MIMD Architectures

• Term partitioning
• When term partitioning is used with an inverted
  file is created for the document collection and
  the inverted lists are spread across the
  processors.
• Assuming each processor has its own I/O channel
  and disks when term distribution in the documents
  and the queries are more skewed, document
  partition performs better. When terms are
  uniformly distributed in user queries, term
  partition performs better.

9
MIMD Architecture
10
SIMD Architecture

• Signature Files

11
SIMD Architecture

• Signature Files

12
SIMD Architecture

• Signature Files

13
SIMD Architectures

• Inverted Files

14
SIMD Architectures
15
SIMD Architectures

• Inverted Files

16
SIMD Architectures
17
Distributed IR

• Introduction
• A distributed computing system can be viewed as a
  MIMD parallel processor with relatively slow
  inter-processor communication channel and the
  freedom to employ a heterogeneous collection of
  processors in the system.

18
Distributed IR

• Introduction
• Distributed Model is very similar to the MIMD
  parallel processing model.
• The main difference here is that subtasks run on
  different computers and the communication between
  the subtasks is performed using network protocol
  such as TCP/IP.

19
Collection Partitioning

• The procedure used to adding documents to search
  servers in a distributed IR system depends a
  number of factors.
• Consider whether or not the system is centrally
  administered.

20
Collection Partitioning

• When the distribute system is centrally
  administered, more options are available.
• The first option is simple replication of the
  collection across all of the search servers.
• The second option is random distribution of the
  documents.
• The final option is explicit semantic
  partitioning of the documents.

21
Source Selection

• Source selection is the process of determining
  which of the distributed document collections are
  most likely to contain relevant documents for the
  current query, and therefore should receive the
  query for processing.
• The basic technique is to treat each collection
  as if it were a single large document, index the
  collections, and evaluate the query against the
  collections to produce a ranked listing of
  collections.

22
Query Processing

• Query processing in a distributed IR system
  proceeds
• as follows
• Select collection to search.
• Distribute query to selected collections.
• Evaluate query at distributed collection in
  parallel.
• Combine results from distributed collection into
  final result.

23
Web Issues

• The parallel and distributed techniques described
  above can then be used directly as if the Web
  were any other large document collection. This is
  the approach currently taken by most of the
  popular Web search services.

24
Trends and Research Issues

• The trend in parallel hardware is the develop of
  general MIMD machines.
• Many challenges remain in the area of parallel
  and distributed text retrieval.
• The first challenge is measuring retrieval
  effectiveness on large text collections.
• The second significant challenge is
  interoperability, or building distributed IR
  systems form heterogeneous components.