My machine continues to encounter the overcaching issue where there is a flood of CPU-only WUs uploaded by the WGG-server to the client every time a GPU-WU is involved in a sync-session. This was an issue in Windows then, and it seems unresolved in Linux now. This issue seems to be a WCG-server issue. The server appears to be blind as to how much of a WU-workload a client can take -- probably the same blindness that causes a rush-WU to interrupt a running-WU even with that running-WU having zero seconds left to complete.

My BOINC is currently in a 'panic mode' as a result of the overcaching-issue with running-WUs all in 'Running, high-priority'.

update1_2012.12.13Th.1242utc:
I calculated the estimated time-to-complete for my remaining WUs (which are all CPU-only WUs) and the resulting projected-date falls at least 2-days before the BOINC-indicated deadline, yet my running WUs still has the RHP ('Running high priority') status. I think that RHP is in error and while that error is in place, I can't get any more WUs even after hitting the 'Update' button several time spaced several minutes apart.
;
; andzgridPost#722
;

You don't list the version of the client you're using but I'll wager a few cool ones that it is 6.x.xx or less.

If you have, or turn on, the flags: <sched_ops>1</sched_ops> and <sched_op_debug>1</sched_op_debug> in the cc_config.xml file you will most likely see in the Event log that everytime the client reports a task complete it asks for a few seconds of CPU work. That's why the cache loads up and everything starts running in "panic" mode.

I found 2 ways to fix the problem, 1) upgrade to version 7.0.28 or later or; 2) set the profile Workunit Cache Settings to "connect every... = 0" and "Cache = 0".

If you would like assistance on how to upgrade the Linux client post your Linux disto, current BOINC client version and how you installed it. Either myself or one of the other members will try to help.

I thought that 7.08 was for Linux and Mac, not windows? I'm seeing my cache reduced on my Windows app_info.xml controlled crunchers but what I'm seeing is a lot of are "Tasks are committed to other platforms" messages, but they are still pulling down work and I haven't seen anything in the results status that indicates I'm rejecting 7.08 tasks.

It *is* GPU 7.08 for Linux and 7.08 CPU for Mac. There's no GPU-Mac yet. The tasks have no specific application version **, the work is a pool. The science application is independent of that. Rest assured that the server wont be trying to send your windows machine a 7.08 application. :D

My Linux machine is still processing pulling new HCC1 for 7.05 [a 64 bit build btw], last one 12 minutes ago . For a moment I thought that the techs wanted the GPU and CPU to have the same version number as what they did on Windows [something to do with deleting and re-fetching the science app [mostly for repair assignments]. The 7.0.xx client does not really care... it's a 6.xx problem.

As for the cache overrun, I posted about this, search and thou shall find, the reason why the mothership project found it necessary to size up the tasks 4 fold, yes 4 fold. The description is, that [old?] clients are deemed too fast, so serial requests are not synching with the server. The mitigation is to get the latest stable. 7.0.38/38 from the Locutusofborg ppa has been doing well for me on Ubuntu. Not seen any overcaching in any form, yet. Of course, 0.0 on MinB/MaxAB works fine too. The 7.0.xx client is at that setting trained to still request minimal work prior to a task finishing, so it does not have idle cores even for a second.

edit: ** only check that is done is, that if copy _0 goes to a version A+platform X, the wingman also goes to version A+platform X. Client 6 did not know science app distiction, just takes the latest, where client 7 does.