In the case you are submitting hundreds of jobs at the same time and your jobs are reading/writing files located at NFS servers, we recommend to limit the number of concurrent running jobs and in this way avoid problems with overload and slow response of NFS servers. The default value has been set to 1000 units per one NFS server.
Add a requirement concurrency_limits into a job_submission_file. Use following syntax and choose the appropriate NFS server (nfs17-25). A job usage of NFS server („consuming of units“) should be also defined. For the first step we recommend to set this value to 1 (1000 / 1 = max 1000 jobs).
Example:
concurrency_limits = nfs18:1If jobs require more intensive access to a NFS server, the value might be increased to e.g. 5 (1000 / 5 = max 200 jobs) or 10 (1000 / 10 = max 100 jobs).
Example:
concurrency_limits = nfs19:5
English 
Čeština