From WikiOAR

Place for Performance Evaluation 2.4's information

Objectives

• Test OAR 2.4 series
• Test scalability upto several thousands of resources (via virtualized infrasctucture)
• Identify performance limitation and propose solution or workaround
• Test alternative systems: SLURM, SGE, Torque...
• Examples of the use of Kameleon to produce G5K environnement
• Try to obtain reproducible experiments

Fiche d'experience

• Objectif
• Procedure
  • environement
  • conditions
• Resultats
• Discussions
• Conclusion
• Logs

Related Works

1. Scalability analysis of job scheduling using virtual nodes Norman Bobroff, Richard Coppinger, Liana Fong, Seetharami Seelam, and Jing Xu

in proceedings of: 14th Workshop on Job Scheduling Strategies for Parallel Processing IPDPS 2009

2. Performance Evaluation of Selected Job Management Systems Kris Gaj and Nguyen Nguyen and Jacek R. Radzikowski and Tarek El-Ghazawi and Nikitas Alexandridis and Frederic Vroman and Preeyapong Samipagdi and Suboh A. Suboh

in proceedings of: Parallel and Distributed Processing Symposium, International IPDPS 2002

3. A Performance Study of Job Management Systems Kris Gaj and Nguyen Nguyen and Jacek R. Radzikowski and Tarek El-Ghazawi and Nikitas Alexandridis and Frederic Vroman and Preeyapong Samipagdi and Suboh A. Suboh

in journal: Concurrency and Computation: Practice & Experience 2004 Systems Performance Evaluation

4. Comparative Study of Distributed Resource Management Systems - SGE, LSF, PBS Pro and Loadleveler Yonghong Yan, Barbara Chapman

Roadmap

Environments

Scheduler Only Evaluation

To evaluate schedulers we follow the sequence below:

1. Populate database: resources, jobs to schedule, previous scheduled jobs if any
2. Launch scheduler
3. Update visualisation for visual control if needed
4. Get start time of last scheduled job for assessment purpose.

Basic OAR Evaluation

Basic Alternative Systems Evaluations