How about if BOINC were to check the currently running WU before it switched to another project/WU?

For example, if the current WU has, let's say, less than 5 minutes to go before it would be completed, BOINC would delay the switch until the current WU completes.

I have on one machine right now a SETI and a LHC WU both with less than 30 seconds for completion!


<a><img>[/url]
<a><img>[/url]

BOINC should wait for either a checkpoint or a WU complete so that it will not lose work if removed from memory.

Would be a good idea if all projects would have short work units.

But imagine, the client program would have to wait for a Climate work unit everytime it wants to switch to a different project.

> Would be a good idea if all projects would have short work units.
>
> But imagine, the client program would have to wait for a Climate work unit
> everytime it wants to switch to a different project.
>
Which is why I said checkpoint or WU complete. Checkpoints happen more frequently than WU complete. The write to disk no more frequently than X sets the minimum time from one checkpoint to the next. The project application determines when it is a good time to ask if BOINC will allow a checkpoint. CPDN asks once every 144 timesteps (3 model days). Predictor Charmm WUs do not have checkpoints (yet, they are being implemented). Other project applications have different timings for checkpoints. On a fast machine checkpoints should be requested every few minutes.

Sorry, I commented the first entry in the thread, I should have mentioned. The checkpoint thing works of course.

> Sorry, I commented the first entry in the thread, I should have mentioned. The
> checkpoint thing works of course.
>
Sorry, it is sometimes hard to tell what the response was to. Unfortunately, waiting for a checkpoint is not the way it currently works. However the Dev team is aware of the problem, and I believe that they are planning a fix for it sometime.