6.6.18

Aus BC-Wiki
Zur Navigation springen Zur Suche springen

Zur Übersicht

  • client: work fetch: in RR sim, keep track of the number of device instances used by jobs that miss deadline. Don't do "variety" work fetch if this is >= # of instances
  • client: reduce frequency of writing state file:
       1) multiply checkpoint period by # of cores (or GPUs)
       2) don't write statefile after schedule_cpus()(we write it after enforce_schedule() if anything started or stopped, and that's all that matters)
  • client: add new log flag <statefile_debug>; tells you when and why statefile is written
  • client: add --no_priority_change cmdline arg (and <no_priority_change> flag in cc_config.xml). If set, run apps at same priority as client.
  • client: garbage collect after scheduler RPC; if project sent some irrelevant FILE_INFOs, this will avoid starting transfers for them.
  • client (linux): if don't find libcudart.so in the current dir, look for it in the library search path. Fixes #863
  • client: give highest work-fetch priority to resources with idle instances
  • client: clear resource backoffs on user-requested RPC
  • client: randomize resource backoffs to avoid lockstep
  • client: when a job finishes, clear the project's backoff for its resource type. This fixes a problem where a project has a "max jobs in progress" limit, and we're backed off because of that. We'll now fetch work immediately instead of waiting 24 hrs.
  • client: handle power events on Windows via a seperate thread and hidden window. The basic code was already in place to handle logoff events on Win9x but now it is enabled for all Windows versions