Scenario 188

Creator HeatForScience
When 28 Jan 2021
Description BOINC Client 7.16.11 (And 7.16.14 on OS X) schedule too many cores (Example has tasks (1 GPU + 1CPU and 12 CPU Tasks) on 12 core system. On systems with more than one GPU, this intermittently results in a GPU task being skipped with an error like: "1/27/2021 4:53:58 PM | collatz | [cpu_sched_debug] skipping GPU job collatz_sieve_4404021a-de64-41b2-bbb3-0b8228e66814_0; CPU committed " Workaround that mostly works in my case: Add app_config.xml with project_max_concurrent to limit CPU tasks for any project enough let the GPU tasks get scheduled first. The CPU tasks then typically end up getting skipped instead. (I add project_max_concurrent to Rosetta and world community grid projects, but this isn't a perfect workaround).
Input files client_state.xml
global_prefs.xml
global_prefs_override.xml
cc_config.xml
Do new simulation

Simulations

ID
Click for details
Who When Parameters Results Comments
1 CJ 5 Mar 2021
Existing jobs only: no
Round-robin only: no
scheduler EDF sim: no
Include empty projects: no
REC half-life: 864000.000000
Simulation duration: 86400.000000
Time step: 60.000000

        
wasted fraction 0.000000
Idle fraction 0.041088
Share violation 1.350763
Monotony 0.663317
RPCs per job 0.093168

        
0 HeatForScience 28 Jan 2021
Existing jobs only: no
Round-robin only: no
scheduler EDF sim: no
Include empty projects: no
REC half-life: 864000.000000
Simulation duration: 86400.000000
Time step: 60.000000

        
wasted fraction 0.000000
Idle fraction 0.041088
Share violation 1.350763
Monotony 0.663317
RPCs per job 0.093168

        


Copyright © 2021 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.