I run 4 projects on an 8 processors computer, (QMC, Einstein, LHC and T4T) each one with the same percentage of cpu. T4T uses only one cpu. Until a few weeks, the scheduler used to share the 7 remaining cpu's between the 3 others projects, 2+2+3, changing every hour. Since 2 or 3 weeks, the scheduler gives all 7 cpu's for only one project, changing project after one or two days. And recently, it no more changes and works only for QMC. It has'nt fetched work for Einstein since 4 days (LHC has no work to send actually). It looks like a system with counter-reaction that becomes unstable and goes to saturation.

I'm running boinc_6.12.34_x86_64-pc-linux-gnu since october 7th

You may think so, but I'm not as cool as you, and since there is written in my preferences "switch between applications every 60 minutes", I think the scheduler has to do it. I really think that the scheduler's behaviour is not was is expected.

still having problems? According to http://einstein.phys.uwm.edu/results.php?hostid=3153307 you have 8 einstein wus in progress, http://qah.uni-muenster.de/results.php?hostid=174054 8 QMC wus in progress, and http://lhcathome2.cern.ch/test4theory/results.php?hostid=9783 1 T4T wu in progress.

Also according to this post http://boinc.berkeley.edu/dev/forum_thread.php?id=6713&nowrap=true#41230 it looks like einstein had an outage. LHC work is extremely patchy, so excluding T4T you are effectively subscribed to only two projects, if one has a hiccup boinc can only fetch from the remaining project. If this has occurred then the project that remained up will have received too much of your cpu time so once the problem project comes back online its likely boinc will fetch more work from that project to try to rectify the balance.

---

CIC1=CC=C(C2=N[C@@H](CC(OC(C)(C)C)=O)C3=NN=C(C)N3C4=C2C(C)=C(C)S4)C=C1

Yesterday, I modified the status of QMC on my computer to "fetch no more job", to see what will happen: as soon as the first QMC wu was ended, Boinc fetched 7 wus from Einstein, put the 6 QMC remaining in waiting state and began to run the 7 Einstein. Why not end first the 6 QMC, in chronological order? I don't understand what logic is behind this behaviour, but if you say it's "normal", well, let it be so.

Not that it's BOINC Manager that does any scheduling. BOINC Manager is the GUI, the graphical command center allowing you to give commands to the client BOINC, which does all the work including the scheduling.

The 6.12 scheduler is finite. It won't be changed anymore. If any big show stopping bugs are in it, too bad. The completely from the ground up rewritten scheduler for BOINC 7, being tested in BOINC 6.13 is what is going to be the next scheduler, which will keep the developers busy to test with at no end.

---