Improved Offsite Running
Try our nova production jobs and various offsite clusters, see what works and what doesn't.
#1 Updated by Enrique Arrieta Diaz about 5 years ago
- File Performance.png added
Offsite locations performance
The performance of the offsite sites is represented by a performance score that runs continuously from 1 to 16, where 1 is the best possible performance. This score takes into account:
- Success Rate, S: number of completed jobs / number of submitted jobs.
- job_time, J: the time elapsed between the first job starts and the end of the last job.
- idle_time< I: the time elapsed between submission of the jobs and the start of the first job.
- Average Time Per File, A.
Score = (S+J+I+A)/4.
The highest success rate gets a 1 and the lowest gets a 16. The lowest: job time, idle time, and average time per file get a 1, and the highest get a 16. If two or more sites tie in their positions they are assigned the same number.
Sites with the lowest performance scores are recommended.
The site named: Offsite, represents the jobs sent offsite using the option: _--offsite_only.
Implementing the offsite locations performance measure is a work in progress.
#18 Updated by Enrique Arrieta Diaz about 5 years ago
The first test included in the plots used the mccheckoutjob.fcl, and 2GB of requested memory. The average time per file was 2 minutes and 12 seconds.
The second and third tests included in the plots used mccheckoutjob.fcl, and 2.4GB of requested memory. The average time per file was 2 minutes and 13 seconds.
The forth test included in the plots used prod_reco_pidpart_numi_job.fcl, and 2GB of requested memory. The average time per file was 126 minutes.