I have recently had a couple of cases where a WU downloaded later than another WU has a deadline before the other WU. As a result, the WU downloaded later preempted the one that was downloaded earlier, so that the first WU downloaded will likely miss out on the "early return" bonus. I'm using BOINC 6.6.17 - it seems BOINC 6.6.x processes WUs in deadline order rather than in download order.

For example, I am currently processing the following WUs:

eZ20842-SH2_US_8-4-10-SH2_US_8840000-1
Downloaded on 31/03/2009 at 11:50:44 AM Deadline 05/04/2009 10:49:36 AM

up854284-pYEpYV_1_US2350000-5-10-ignasi_0
Downloaded on 31/03/2009 at 11:08:16 PM Deadline 04/04/2009 11:07:13 PM

This is running on a 9800GT which typically takes around 12-14 hours to process a WU.

Any idea why this might be happening?

Sure.

Note that GDF says that "new WUs will have deadline of 5 days, starting last Friday". Your second WU is previous to all this therefore it continues with the old deadline. Whereas all SH2_US_8 are already 5 days deadline.

All old WUs will be clearing out as they finish their workflows.

Just a bit of patience.
thanks.

Second WU was downloaded on Tuesday, which is later than Friday.

But I will be patient!

Because they are continuations of batches sent before Friday.

Simulations are in fact much longer than what a single card can do in a reasonable (and volunteer-friendly) amount of time, something between 6 to 14 hours (non-stop and depending on the GPU) let's say. So what we do is split the time length of the simulations in "steps", so you might compute the first step, then the second (exact continuation of your simulation) is sent to someone else, then the third and so on... until 10, which is what we are setting things up to lately.

So until all old WU "workflows" do reach their final step, they'll have the old setup deadline of 4, unlike the new ones that are sent with a deadline of 5 days.

thanks,
ignasi

ok thanks ignasi. Seem to be getting all "new" 5 day WUs now.

Simulations are in fact much longer than what a single card can do in a reasonable (and volunteer-friendly) amount of time, something between 6 to 14 hours (non-stop and depending on the GPU) let's say. So what we do is split the time length of the simulations in "steps", so you might compute the first step, then the second (exact continuation of your simulation) is sent to someone else, then the third and so on... until 10, which is what we are setting things up to lately.