Analysis of Overhead and Waiting Time in the EGEE Production Grid Proceedings of the Cracow Grid Workshop 2008 Maximilian Berger Thomas Zangerl Thomas Fahringer In the paper we present an analysis of the EGEE production Grid infrastructure, to verify if it fulfills the promised claim of providing a scalable computing resource plattform. We measured scheduling latency and information service overhead. In our measurements we discovered that the speed of the gLite middleware has significantly improved in the last two years. We have taken measurements for every step of the job lifecycle. We have also measured the actual time when a job started and finished its execution through a callback-mechanism. We present current results which show that the information service introduces an an additional overhead between the time a job actually finishes and the notification of the user. We also analyzed the the delays over weekdays and hours of the day. Where did not find any significant results which would support the "weekend effect". We discovered that the job latency and reliability are directly dependent on the site a job is actually scheduled to, and that some sites seem to be much faster and more reliable than others. We conclude that a simple analysis of scheduling time in the EGEE network or on one single site does not provide sufficient results – the factor site plays a large role in the actual scheduling latency. We outline some of the possible decision changes that could be made to improve scheduling in the EGEE infrastructure. We also set the overhead time in relation with job run-time, to estimate the types of jobs suitable for EGEE Grid execution. http://max.berger.name/research/analysis_time_egee.pdf 978-83-61433-00-2 CGW 08: Cracow Grid Workshop Oct 13-15, 2008

Crakow Poland

287-294 2009