MOO Wrapper Another unexpected outage

[RSS]

There was another out-of-memory event that affected our main (and only) server to bring the project down for about six hours last Friday. Services were fully down from 21:00 to 3:30 local time (EEST+3, so that's 18:00 to 0:30 UTC and 11:00 to 17:30 PDT).Anybody who is interested reading the long technical details, please see the http://moowrap.net/forum_thread.php?id=113 forum post. Thanks for crunching!

More... (http://moowrap.net/forum_thread.php?id=114)

[zombie67]

I'll use this as an opportunity to talk about a feature in BOINC that will prevent downtime for us crunchers.

Lets say that there is down time for your fav GPU project. You would have to monitor your machines constantly to find out about it. There a couple of alternatives:

1) Normal resource share: This solution is to have a high resource share (say 200) for your fav GPU project, with a low resource share (say 5) for your backup project. But this means that you don't get 100% of your GPU applied to your fav during up time. It also doesn't work well when your fav project has a limited queue setting (your queue fills up with your back up project greater than normal resource share would require).

2) "0" resource share: The other option is to set your backup project with a "0" resource share. This requires a modern BOINC client, and a modern BOINC server. I am not sure of the versions of each. Look it up on trac if you are curious. In any case, when your GPU project with a resource share of >0 runs out of work, or is down, or whatever, BOINC will ask for work from your backup project with a resource share of "0". And it will download only enough work to keep the GPUs busy right not. It will not fill up the queue. It will continue with this behavior until your fav has work again. The benefit here is that you don't have to monitor your machines, or take any action when your fav is down. Effortless and painless.

I recommend all GPU crunchers use #2.

[Bryan]

I'm running 6.10.58 and .60 and if I set Collatz up as the backup it still loads the cache. I haven't migrated to the newer versions of BOINC so they may have fixed the problem.

[zombie67]

I'm running 6.10.58 and .60 and if I set Collatz up as the backup it still loads the cache. I haven't migrated to the newer versions of BOINC so they may have fixed the problem.

That may or may not be the client's issue. Like I said, It ALSO requires the BOINC server to be of a certain version. FWIW, 6.10.58/60 do have the feature. And it works fine for me.

[Bryan]

I've used it on some projects and it worked very well ... finished a wu and then downloaded the next. A long time ago Slicker said there was a problem in BOINC (client I believe) that caused it to screw up on Collatz. I put Collatz as my backup when DirT was running dry yesterday and it downloaded about 100 wu when it started.

It is really a neat feature especially when it works as intended! I noticed that MW now also allows a "0" setting. It didn't used to do that so that is an improvement. Of course since I haven't actually run MW in a year I guess they could have updated to allow it a very long time ago :))

[NJCaNS]

Nice tip Z. My 2 cents. Won't this tip get lost/buried on this thread?

Do we have an FAQ section or a post section for Tip, Tricks or GPU crunching where this tip might become more useful (or found more often)? Admin's?