Outline

Motivations

Mobile Assistant Programming model

Implementation: MAP Scheme for WWW

Application

Conclusions

Motivation

Synchronous access

• limits performance
• is not suitable for disconnected operation

New applications

• resource discovery
• data mining
• agent programming

Application example

New paradigm

Asynchronous activation and result retrieving

Moving computation to data

Transparent handling of failures and disconnections

Exploitation of massive parallelism

Interpreted code running on heterogeneous platforms

MAP: Mobile Assistant Programming

Provide flexible mechanisms for programming applications accessing information on the WWW:

• asynchronous operation
• persistent processes that evolve in the network
• parallel execution
• high-level interpreted programs

Principles of the model

MAP architecture

Assistant primitives

• creates an assistant to execute source-code on activation-node and specifies result-node to collect result reports in collector-id

• gets the results from collector-id

MAP primitives

• moves an assistant to a remote node

• clones an assistant

• reports results to the collector object

• stops execution

Implementation

Scheme

• fully fledged programming language
• first-class procedures
• interpreted

WWW

• large scale information space
• fast prototyping - HTTP Post method, CGI scripts

Implementation architecture

Implementation

Modified Scheme interpreter

• unsafe functions removed (open, read, write)
• MAP primitives added
• WWW access primitives (get, parse)

System support

• save and restore execution state
• transfer execution state

Information access application

Search for HTML documents on a set of WWW

• query
• DNS domain
• root URL
• regular expression for the name of WWW servers

Initial form

text:   
title:  
keyword:
author: 

Result collector page

WebFinder

• MIT LCS
• IMAG LSR

To obtain partial or total results click here

Result page

WebFinder

Results to query ( text : research and keyword : network ) are

• MIT LCS
  • MIT Lab for Computer Science Web Page (6423Ko)
  • LCS Advanced Network Architecture Papers (21956Ko)
  • COMPUTER ARCHITECTURE GROUP (4625Ko)
  • Other ANA Group Publications (11965Ko)
• IMAG LSR

Performance

MAP vs. robot

• MAP: analyze 121 pages of 1300KB; transfer 150KB of results
• robot: transfer all documents; analyze locally

Elapsed time

• MAP: 540 s
• robot: 2277s

Conclusions

New model for efficient information access on the WWW

Current and future work

• implementation of an authentication scheme
• control of resource consumption

Platform deployment

• if you want to use MAP, install MapServer and use all servers in the pool