The rule is that a client tries to return a task 24 hours before deadline, thus if received one say 40 hours before deadline and it reckons it's taking 20 hours to complete, it will kick in the EDF rule before you can even blink but no later than that 24 hour "before" target return time is under threat.

Anytime you suspend a task to push another forward you cause a lot of whirring, plus work fetching is suspended until all tasks are released again, but in your case you seem to be doing that immediately.

If you want to push a task you better make sure to first do the once you dont want to push in the Ready to Start condition before suspending jobs already running or you could end up with tasks that hog memory with 0, 1 or a few seconds on the clock. BOINC processes multi line commands sequentially from the top in the order they have been sorted on display.

A good observer had seen that these beta's had an unexpected wide spread and those same observers had read knreed's post explaining that the project will track and compute the 3 types so it can tell BOINC what to expect in fpops from which it derives the run-times.

In fact it depends. The Duration Correction Factor (DCF) is adjusted every time a WU finishes. So if you are running HCC only on a quad you might be back to a more reasonable estimate within one day or two. Or even faster if it is an 8-core.
Unless you get a new repair Beta WU as I have just had, with the usual unreachable deadline (no worry, see other threads)...

Cheers. Jean.

A suggestion was made earlier in the back room v.v. this condition. There's a reasonable probability that the extra copies come back after the "No reply" returns evading a "Too Late", not sure. Just let it crunch and use the opportunity to increase chance for that coveted Beta badge.

Please note that you will get credit for a late result as long as it is returned within 5 days of the deadline.

All results are useful to us during beta testing as we are looking for ratio's between valid and invalid and successes and errors.

So please allow them to finish running.

We are also looking for data on how long the workunits actually take and what variation exists between different workunits.

As a reference, we can run more work through beta in one day then we can in many weeks of our 'alpha' testing. Additionally, our alpha environment tends to have somewhat homogeneous computers compared to the diversity of computers that you all have.

The beta testing is an extremely valuable tool to us. But it is still testing and there will be oddities (oversized workunits, bad flops estimates, short deadlines, etc). Some of you might remember back before we had beta testing and these problems would always appear shortly after launch on many computers of people who are not as dedicate as you folks. This was a much bigger problem.

We definitely appreciate all of your comments and help - please keep them coming!