On 15 Mar 2016, at 11:45, Simon Fayer <simon.fayer05@IMPERIAL.AC.UK> wrote:
Hi Steve,
It looks like you're using the whole time that the job is in the "Running" state from the logging info. This includes the sandbox stage-in time (but bizarrely not the stage-out time), so it is probably just a slow network link into the WN/VM network. (For more accurate user payload timing it's best to only measure the "Job_<number> | Running" states).
There isn't much more we can investigate about this as we don't get any pilot logs from VAC/vcycle sites (I don't know whether there is any plan to make these available in future versions, if such a log even exists on these nodes).
There’s a quick hack for the VMs at the moment: https://depo.gridpp.ac.uk/gridpp-vm.tier2.hep.manchester.ac.uk/vcycle-cern.t... and The Plan is for DIRAC to have a general pilot log file service which will give you the output of the DIRAC pilot scripts. This will also address it for batch platforms where DIRAC doesn’t currently retrieve the pilot logs through the batch API (e.g. HTCondorCE.) Cheers Andrew -- Dr Andrew McNab University of Manchester High Energy Physics, LHCb@CERN (Distributed Computing Coordinator), and GridPP (LHCb + Tier-2 Evolution) www.hep.manchester.ac.uk/u/mcnab Skype: andrew.mcnab.uk