Seems there is no control on how many jobs an host can cache

[rdemaria]

Comptuing options in the client allow to specify Store at least n days of work and Store up to an additional n days of work. However in the server we have the options

 <max_wus_in_progress> 2 </max_wus_in_progress>

thats is multiplied by the ncpu.

We still observe several hosts that have many more that 2 tasks per cpus. This can be reproduced by increasing to 1 Store at least n days of work and Store up to an additional n days of work in a machine with 1 CPU.

[rdemaria]

See also config_aux.html in https://boinc.berkeley.edu/trac/wiki/ProjectOptions. Maybe this works...

[rdemaria]

https://cern.service-now.com/service-portal/view-request.do?n=RQF0806520

[Toby-Broom]

My computers will buffer alot of work even if I have store 0.5day and 0.01 day extra. I think the flops estimate is used in the calculations?

[rdemaria]

I am also experiencing the same issue in my clients. I don't know the cause. We were thinking to address this by decreasing the timeout time, such that pending tasks are removed.

[Toby-Broom]

Maybe you could ask David Anderson?, seems to me the setting you used should have worked? I see now that I have 75 tasks on a 4 core machine this should have only 8? For Theory they lowered the max_ncpus and it worked as my hosts only got 10 WU even when I had many unused cores, so I asked for it to be removed. Mabey this has to be set also to enable the limits? then I would expect the max number of WU to be 128, I think I have seen more during the big submission a while back but I don't know if the server setting were enabled then?

[amereghe]

I suspect it is related to a wrong fpops_estimate, which is then used, together with the peak CPU FLOPs, to estimate the time required for each task... Toby, what is the expected time for the tasks that you receive? Could you report a couple of cases together with their names?

[Toby-Broom]

Workunit name Dtwo_70_hlbbo_2222_1.1_0.75__1__s__62.309_60.3119__2_4__6__81_1_sixvf_boinc3628
Estimated app speed 5.60 GFLOPs/sec
Estimated task size 180,000 GFLOPs
CPU time 01:02:00
Estimated time remaining 07:27:21

Workunit name w-c0_job_helhc_n10060__59__s__62.28_60.31__2_3__6__49.5_1_sixvf_boinc14665
Estimated app speed 11.34 GFLOPs/sec
Estimated task size 180,000 GFLOPs
CPU time 02:11:59
Elapsed time 02:12:38
Estimated time remaining 00:18:32

[Toby-Broom]

These aren't from the 75 task PC as they were all crunched this morning.