Starting new tasks before old ones are done.

[orion456]

A few of my machines have begun to start new tasks before the old ones are finished. In some cases they are 75% finished and they simply stop being updated and a new task is started. Sometimes I might have 10 tasks started, with only a few actually being processed.

Is this some kind of bug or what?? Perhaps a new feature!! :bang head

[xrror]

This has happened for a long time now (ever since the BOINC 5.x clients at least), I have no idea why - and yes it is super annoying.

I too would love to know why this happens. My completely uneducated non-fact based guess is it's something with the BOINC scheduler getting "creative" with rotating projects...

...but I only run 1 project, and I put it at 100% rosetta. And it still happens. :confused:

The worst is if you run a large buffer, and it does this so much that work units start expiring, and they're 70%+ done when they're killed :bang head

Even if they don't expire, it gets old when a machine starts "sand-bagging" where 3 days of work "builds up" and then it's all suddenly finished. Really messes with the stats and makes monitoring machines remotely a PITA. A machine will show zero credit for 3 days, then suddenly dump like 15000 RAC.

Hopefully someone has more info on this.

[orion456]

Good to know I'm not alone in this problem.

Is there some central site where such things are discussed in detail?

[orion456]

The apparent fix for this problem is to simply reboot. Somehow BOINC gets confused every so often and a good reboot causes it to return to normal behavior after using up all the incorrectly started instances.

[LiGhTBoY]

This has happened for a long time now (ever since the BOINC 5.x clients at least), I have no idea why - and yes it is super annoying.

I too would love to know why this happens. My completely uneducated non-fact based guess is it's something with the BOINC scheduler getting "creative" with rotating projects...

...but I only run 1 project, and I put it at 100% rosetta. And it still happens. :confused:

The worst is if you run a large buffer, and it does this so much that work units start expiring, and they're 70%+ done when they're killed :bang head

Even if they don't expire, it gets old when a machine starts "sand-bagging" where 3 days of work "builds up" and then it's all suddenly finished. Really messes with the stats and makes monitoring machines remotely a PITA. A machine will show zero credit for 3 days, then suddenly dump like 15000 RAC.

Hopefully someone has more info on this.

I have noticed the exact same thing. The way to fix this problem is to reduce your work buffer. Once reduced, the problem goes away, until you either "overload" the pc in a different way (gaming for hours) or you turn it off.

As someone else mentioned, it's the scheduler getting "creative", but a feature to control how to allocate the CPU time to the workunits based on their percentage done would be great.