Download presentation
Presentation is loading. Please wait.
Published byMaximilian O’Neal’ Modified over 8 years ago
1
CSS497 Undergraduate Research Performance Comparison Among Agent Teamwork, Globus and Condor By Timothy Chuang Advisor: Professor Munehiro Fukuda
2
Overview Agent Teamwork – deployment of mobile agents Agents launch, monitor and resume jobs Fault-tolerant Condor – opportunist job dispatcher Condor daemon searches for idle computing nodes on which to dispatch jobs Emphasize on job migration upon encountering an error Globus – widely used grid computing middleware MPICH is required for parallel applications
3
Condor User Condor Pool X Gateway Class Manager Snapshot Class Manager
4
Globus LFSPBSGRAMs DUROC/MPICH-G2 User
5
Agent Teamwork FTP Server User A User B User B snapshot snapshots User program wrapper Snapshot Methods GridTCP User program wrapper Snapshot Methods GridTCP User program wrapper Snapshot Methods GridTCP snapshot User A’s Process User A’s Process User B’s Process TCP Communication Commander Agent Sentinel Agent Resource Agent Sentinel Agent Resource Agent Bookkeeper Agent Results
6
Project Objectives Establish reference platform Condor Installation PVM installation Implement parallel applications to run on PVM Matrix Multiplication Wave2D Simulation Mandelbrot Set Simulation Distributed Grep
7
Modify parallel the same applications to utilize Agent Teamwork’s check pointing feature Check previous Globus status Convert the same parallel applications to MPICH-G2 Conduct performance evaluation
8
Problems with Condor/PVM Condor no longer fully Supports PVM PVM universe to dispatch jobs in is no longer functional As a result, condor was dropped from the project
9
Evaluation of Agent Teamwork’s Fault-tolerance Performance Applications used Matrix Multiplication Mandelbrot Set Renderer Wave2D Simulation Distributed Grep Fault-tolerance Performance Evaluate the extra overhead of checkpointing and resumption
10
Challenges Finding a large problem set that can scale well with the increasing number of computing nodes Certain problem sizes are limited to the master node’s memory – Matrix Multiplication Debugging parallel applications Requires going through time consuming diagnosis Finding the best check-pointing frequency for all applications Setting the frequency too low could take up to three hours to finish a job!
11
Performance - MatrixMult
12
Performance – Wave2D
13
Performance – Mandelbrot
14
Performance – Distributed Grep
15
Continued Work Scale problem size to utilize all 64 computing nodes Conduct performance evaluation on multi-clusters Conduct performance evaluation on Globus Compare Globus’ performance with Agent Teamwork
16
Useful Classes CSS301 – Technical Writing CSS343 – Data Structures and Algorithms CSS430 – Operating Systems CSS432 – Network Design CSS434 – Parallel and Distributed Computing
17
Acknowledgements My Faculty Advisor: Professor Munehiro Fukuda UWB Linux System Administrators: Mr. David Grimmer Mrs. Meryll Larkin My Sponsor: Mr. Joshua Phillips
18
Questions?
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.