A follow-up to thread id=10389, now locked...

I was curious to see how system resources would be allocated among multiple projects. I set up the resource shares as noted below and let boinc run without any manual intervention for 75 days to aim to reach some steady state. I've captured run time and credit statistics for the 5-day span (Sept. 23 - 27) and tabulated the results below. (It has taken a while for work units to be validated and credits granted.)

project  share  GPU used  CPU used  credit  RAC 9/27
-------   ---    ------    ------    ----    ------
GPU+CPU
  Seti      88 %    85 %      38 %    37 %    5400
  Einstein  11 %    14 %       5 %    23 %    4391
CPU
  NFS        1 %     n.a.     24 %    13 %    1380
  Asteroids  1 %     n.a.     23 %    27 %    3658

The GPU (GTX 650) is configured to run only one task at a time. The CPU has 4 cores, one is reserved/configured to support the GPU. The Einstein, NFS, and Asteroids projects are configured to "max concurrent" = 1. The Seti and Einstein projects seem to share the GPU and the CPU resources close to the 8:1 share ratio settings. The NFS and Asteroids projects are not close. To calculate the % CPU used I have assumed 3-cores were available. So the NFS project, for example, used 24% of the 3 cores but that translates into using its "max concurrent" core 72% of the time over those 5 days.

In the past I have succumbed to the temptation to "micro manage" the work buffer, eg. setting NNT for resource hogs occasionally to allow downloads for other projects with a higher share. (The buffer parameter is set for 1+0.5 day.) As of 8 a.m. today the CPU work buffer for Seti is zero while the NFS and Asteroids projects have 57 and 77 hours respectively. This is admittedly just a snapshot sample but it is not unusual. Seti CPU work has been 0 on 12 of the last 30 days.

The boinc client is 7.4.23 (a Debian Linux distribution). There appears to be a 7.6.12 version in the pipeline and I look forward to upgrading although I have not seen anything in the message boards to suggest any change in the resource share management.

Just my 2-cents worth to present real user statistics and not just anecdotal comments. I wish that boinc did a "better" job of resource sharing but I will "play with the hand I'm dealt."