BeautyDIRAC v1r0 (AKA LHCbDIRAC v7r15) - operational changes

Changes in the production steps management (applies to all application managers)

  • Steps have 2 new parameters: "SystemConfig" and "MC TCK".
    • The system config is the CMT system config the application specified in the step can use. For example, "x86_64-slc5-gcc46-opt".
    • MC application managers have the possibility to specify an "MC TCK" as one of the parameters of the steps.

Changes for all the transformation types (all productions and replications)

  • There is a new status for Productions, named "Idle". It is reached when the production is REALLY doing nothing.
  • The TransformationMonitor web page slightly changed, so please refresh it.

Changes for all the productions:

  • When an LFN is reported as "fail" or "part" in the XML summary, it will be set as "problematic" in the TransformationDB.

Changes for MC productions:

  • The new MC jobs (so, not those that are NOW in the system, but those of the new productions that you will launch) will produce a non-fixed amount of events, but will work using the CPUTimeLeft of the queue, and the CPUNormalizationFactor of the CPU where they land.
    • There is anyway a hard limit, hardcoded, that no MC job will last more than ~24hours of real time.
  • When launching a new MC production, you will have to specify the CPUe (CPU time per event, in HS06s). I have seen that, for fast generations, it can be down to ~300 HS06s, but I suggest to be always a bit generous. This value can be calculated by simply doing CPUTimeOfEvent * CPUNormalizationFactor of the machine, where CPUTimeOfEvent is the real seconds it took generating an event on the test machine. An MC testing machinery will come, but since it has not been yet developed,
    • In the template of MC prods, you won't find any more the MaxCPUTime (for which you usually put 700000 HS06s, I think), but a minimum value. The default is 50000 HS06s, which seems reasonable. This way, almost any queue will match a MC job, just they will produce less events
    • You won't need to specify the number of jobs to create, this will be done automatically
    • You won't need to extend the MC prod, again this will be done automatically by a new agent, the MCExtensionAgent, when the MC production is Idle, that will pick info from the ProductionRequest page.
    • You won't need to specify the SystemConfig, because, as said above, for new steps it is now a parameter of the steps, not of the productions. For productions that use steps already in the system, or new steps that don't specify the SystemConfig, the default will be set to "x86_64-slc5-gcc46-opt".
  • The ValidatingInput and ValidatingOutput status can now be reached ONLY from the Idle status.

-- FedericoStagni - 21 Jan 2014

Edit | Attach | Watch | Print version | History: r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r1 - 2014-01-21 - FedericoStagni
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    LHCb All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright &© 2008-2023 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback