MW work fetch policy

[BF]

I'm not understanding the how BOINC fetches work for MW..

I had 24 completed WU sitting there, every other project on the the computer suspended, all cores sitting idle.  But MW was still not reporting the work and requesting new WUs.  And it was not sitting there in "Communication deferred for x minutes" either.  Just sitting there twiddling its virtual thumbs.

When I hit update it reported the completed WUs, but still did not request more work!  ??? :hbang:

[WikiWill]

Major pain in the rear.  Which version Boinc client?

[yoda]

I've seen the same thing.  No work on hand, yet no work being requested.  Manual updates showed 0 work being requested too

Unfortunately I have no idea on what's causing it. let alone know how to fix it.  In my case, MW was the ONLY project the PC was attached to as well...  it did eventually download work after I reset Milkyway

[WikiWill]

I haven't seen it on MW but I did see it when running Cels@home, both with and without parallel projects.

That was a 6.2.x client version I think although it's probably not a version-specific bug.

The problem was more pronounced on Cels because only one WU downloaded per request.

[Rocoll]

It's an Alpha project and needs another project running as well - even if it's set at 5%, so you don't run out of work.

BF haven't seen what you describe where it doesn't go to the other project that's waiting?

[Wang Solutions]

Yes, I am getting the same thing all the time, so have had to have every computer also running another project besides MW so they don't sit idle. At one point I thought it was because I was attached to GPU Grid and that BOINC thought it had enough work for MW not to request more, but the same thing is happening on computers not attached to GPU Grid.

I believe it is an issue to do with the way the MW server is configured. I believe the admins are aware of the issue but anything they have done seems to have made it worse rather than better. 

[Furlozza]

*cynic rising again*

Of course, it could be part of the program itself to slow down the amount that has to pass through the server.

*cynic falling*

I just bit the bullet and put WCG on a fall back since have been bitten three times at night by MW not downloading work at all. Have noticed that a couple of times when I did restart after above break that MW downloaded a master file.....

[BF]

Major pain in the rear.  Which version Boinc client?

6.6.3 (an out of date dev version).. but it's also happening on another computer on 6.4.7, the recommended release for Win 64bit..

I believe it is an issue to do with the way the MW server is configured. I believe the admins are aware of the issue but anything they have done seems to have made it worse rather than better. 

Ah, that explains it then.. I had a quick look on the MW boards, but couldn't find anything relevant.. will have a more thorough look now..

[clownius]

What are people using as their connect to and work buffer policies?  I just changed mine from connect every .25 days and a buffer of .25 days to connect every 0 days and a buffer of .2 days to see if it helps.

[BF]

I don't think that it will make a difference, it seems to be a problem with the MW app and not BOINC as this doesn't seem to happen with other projects..

Got up this morning to find 3 boxes sitting idle not requesting any work on MW since 3am.  :hbang: :hbang: :hbang:

I'll add some back-up projects, but I really didn't want to do that.

[clownius]

Your right i just dried up again anyways.  MW refuses to give me new tasks.  Ill run ABC as a backup. I was slipping in ABC anyways.

[WikiWill]

Likewise, I twiddled cache and contact settings to no avail.

I set up an hourly scheduled task on my Windows boxes to do:
"C:\Program files\Boinc\boinccmd.exe" --project http://milkywayurl update
And today's CPU log looks much more consistently 100%. Time will tell in the Boincstats I guess.

[kashi]

I think the cause is the combination of the single server and the GPU app, it's too fast!

Single and multi GPU computers would be completing their allocated WUs in 5-10 minutes and frequently requesting more work. I don't know if anyone is running this configuration but in theory a dual core running 3 x HD4870 would be completing its allocation every 100 seconds. This is assuming only larger WUs were allocated, any shorter ones issued and these times are reduced.

Only a small number of people using the GPU app would be equivalent to a very large farm crunching MilkyWay, remember the trouble there was getting work whenever Banshee was active.

[BF]

I think that's a seperate issue Kashi.. I wasn't referring to the fact that MW sometimes has no work available.. My issue is that sometimes BOINC doesn't put out a request for more work in the first place.

As soon as I reset the project on the box (and re-installed the opti app) BOINC requested and received MW work again.. just waiting for it to happen again now grrr...

[kashi]

My apologies, I misunderstood.

It is possible the problem is still related to server overload though. I have noticed that when BOINC requests more work when you already have the maximum allowed it states  "Scheduler request completed: got 0 new tasks" but sometimes fails to give the red "Message from server: No work sent Message from server: (reached per-CPU limit of 6 tasks)". That means it is connecting to the server but the server is not completing its full response. If the overloaded server is not always reponding completely then the reset giving the "Fetching scheduler list Master file download succeeded" message after the cycle of backoffs may not occur. Without the scheduling being reset from the server then BOINC will just sit in a idle state.