Work shortage and previous outages
It seems we run out of work to send today for a second time during this week (last incident happened three days ago) and both of them has been because upstream (distributed.net) keymaster has not send us any work to generate into work units for you to crunch on.
Three days ago the keymaster run out of work (they currently have limited alerts for such issues due to a monitoring server failure) until an operator queued more work and today the keymaster was somehow confused and not sending anything until the same operator restarted the service. So work is getting generated and send to you again.
Also, there has been some total outages earlier that has been due to our primary database server crashing. I'm still trying to figure out what's the exact cause of it's crashing under high load but for now my work a rounds has been keeping it up. If all else fails, I'm going to change server hardware to one that's hopefully more reliable.
21 Mar 2013, 18:51:56 UTC · Comment
Plan class configuration and requirements changed
I'm switching to using an external XML file to define plan classes (for technical details, please read http://boinc.berkeley.edu/trac/wiki/AppPlanSpec) so there can be some strange responses from scheduler while things stabilize. I'll try watch how the scheduler behaves but, please, let me know if your previously working systems no longer get any work assigned.
Additionally, I'm going to lower the ATI memory requirements so that some cards with limited memory can also participate. Native ATI Stream app is no longer available for ATI 7xxx class cards in preparation for getting a working OpenCL app out for them.
Apologies for any problems and thanks for your understanding and crunch power!
15 Dec 2012, 17:57:11 UTC · Comment
Server code update and fix to GPU work scheduling
Just finished an update to the latest BOINC Server code that brought a lot of fixes and improvements from BOINC devs in last 6 months. These included web feature changes as well as various backend process changes. (And before you ask, no, credit system is still the same and no granting adjustments have been done there.)
Additionally, since the code refresh didn't seem to fix it, I debugged the "only 1 GPU work unit sent by scheduler" problem and hopefully got it fixed. It seems the matchmaker scoring scheduling algorithm we switched to some time ago doesn't work correctly for GPU work request. (Specifically the new resource based work requests of new BOINC client versions.) I switched us back to the old trusty array scheduling algorithm and now the GPU work requests are been fulfilled fully by our scheduler.
As always, please report any oddities you might encounter and thanks for crunching!
10 Sep 2012, 4:17:53 UTC · Comment
News is available as an RSS feed