Since the 6.10.58 client I've noticed numerous problems with the scheduler downloading work units and scheduling them.

I have two Win XP SP3 quad core machines with 9800GT's, one Home on an Intel and one Pro on a Phenom. Both had the same issues so it doesn't seem to be an install problem. Also these things didn't happen before 6.10.58 at least that I noticed. Active projects on both are SETI@Home and Einstein@Home. At the time SETI was down so only Einstein had work.

Some examples:

1) The scheduler will sometimes request and receive work far in excess of the cache setting. See this thread.

2) The client seems to get stuck in "GPU mode" or "CPU mode" where it doesn't want to download both types of work, regardless of how much is remaining. It can be forced by setting No New Tasks for the project until the project's timeout is elapsed and work can be requested, then quickly clicking Allow New Tasks and Update.

Example:

07/12/2010 8:37:39 PM	SETI@home	Sending scheduler request: To fetch work.
07/12/2010 8:37:39 PM	SETI@home	Requesting new tasks for GPU
07/12/2010 8:37:41 PM	SETI@home	Scheduler request completed: got 0 new tasks
07/12/2010 8:37:41 PM	SETI@home	Message from server: Project has no tasks available
07/12/2010 8:42:46 PM	SETI@home	Sending scheduler request: To fetch work.
07/12/2010 8:42:46 PM	SETI@home	Requesting new tasks for GPU
07/12/2010 8:42:47 PM	SETI@home	Scheduler request completed: got 0 new tasks
07/12/2010 8:42:47 PM	SETI@home	Message from server: Project has no tasks available
07/12/2010 8:47:17 PM	SETI@home	work fetch suspended by user
07/12/2010 8:47:55 PM	SETI@home	work fetch resumed by user
07/12/2010 8:47:55 PM	SETI@home	update requested by user
07/12/2010 8:47:57 PM	SETI@home	Sending scheduler request: Requested by user.
07/12/2010 8:47:57 PM	SETI@home	Requesting new tasks for CPU and GPU
07/12/2010 8:47:59 PM	SETI@home	Scheduler request completed: got 0 new tasks
07/12/2010 8:47:59 PM	SETI@home	Message from server: Project has no tasks available

The last one was when I suspended work fetch and manually initiated fetch as indicated above. Only then would BOINC request CPU work. This was with a five-day cache and only about half a day of CPU work remaining.

3) BOINC will pre-empt a project with workunits from the same project:



That machine has four cores, so the remainder of these are EAH workunits that were pre-empted by other EAH workunits of exactly the same type. This may be due to receiving units with an older deadline though which would be by design.

---