6.10.29

Aus BC-Wiki
Wechseln zu: Navigation, Suche

Zur Übersicht

  • client: show <cpu_sched> message when suspend/resume GPU activity
  • client: there are oddball cases where the client has an app for which avg_ncpus > ncpus. Allow such jobs to run; otherwise nothing runs.
  • client: in computing "anticipated" STD, scale by ninstances (from John McLeod)
  • MGR: Deconflict menu item IDs.
  • manager/client: show PID in task properties; add PID to RESULT structure in GUI RPC
  • client: when parsing cc_config.xml, clear vectors so we don't end up with duplicates
  • client: divide LTD deltas by ninstances, same as for STD. This is cosmetic - it won't affect work fetch, but it will prevent LTD from changing faster than real time
  • client: add logic to work fetch so that each project will have enough jobs to use its share of resource instances. This avoids situations where e.g. on a 2-CPU system a project has 75% resource share and 1 CPU job, and its STD increases without bound. Did a general cleanup of the logic for computing work request sizes (seconds and instances).
  • client: Unix: if can't create client lock file, report ERR_OPEN rather than ERR_ALREADY_RUNNING fixes #970
  • client: fix format strings for ninstances (can be fraction now)
  • client: scheduling problem:
    • a project overestimates job FLOP counts
    • the client starts jobs in EDF mode
    • as job progresses and fraction done increases, its completion time estimate decreases until it's no longer a deadline miss.
  • job gets preempted by other job from that project; you end up with lots of partly completed jobs. Solution (I hope): if an app version has running jobs, compute a "temp DCF" for the app version, which is the min of dynamic/static estimates for its jobs. Apply this scaling factor to completion time estimates for unstarted jobs in RR simulation
  • client: the estimation of remaining time of running jobs was wrong (how did this bug survive so long?)
  • client: bug in ACTIVE_TASK::est_dur()
  • client: work fetch fix: avoid sending null request in certain cases.
  • client: add <stderr_head> config option; sends the first rather than last 64KB of stderr to server. This doesn't belong here; this choice should come from the server. I may take this out later.
  • client: small tweak to work fetch: if project has crazy DCF, don't automatically request 1 sec; only request work if there's a shortfall.