It's happened quite a few more times than you would think. A WU have less than 10 minutes before being complete, and then BOINC change the active project according to sharefactor and the "change application every x minutes" setting. This is just folly! I'm 'only' running three projects, but still it's a matter of HOURS before boinc returns to the WU nearly finished.

When this is brought up, it's also necessary to bring up, that the seti-application ought to be better at calculating the remaining as opposed to now, where it few seconds after starting a WU thinks that about 30 minutes of computing remains. It then flipflops and start counting UPWARDS for for a good 10 minutes before starting to count down again! and not only that; the the remaining time stays at the same point for several seconds before counting one second down. What is it in the workings of SAH that makes this time-calculation so screwy and unreliable??

---

[br]BOINC is A.W.E.S.O.M.E, and then some.[br]

I agree that there should be a better analysis at the time remaining prior to swapping out a project.

I asume that the time remaining must, at some point, actually count down because the WU do get completed, but like Peter, I have only seen mine count up.

---

...don

There is a task in the taskbase to wait for the next checkpoint or a WU complete after the expiration of the time slice. This should solve the problem.

---

BOINC WIKI

The tendency for time remaining to increase is due to the dynamic nature of WU processing in S@H. The more interesting a "signal" appears to be, the more time is spent analyzing it. Work units containing a lot of noise may take a very long time to process.

---

> The tendency for time remaining to increase is due to the dynamic nature of WU
> processing in S@H. The more interesting a "signal" appears to be, the more
> time is spent analyzing it. Work units containing a lot of noise may take a
> very long time to process.
>
But if they contain too much noise, they are aborted almost immediately ~5min. Still, BOINC should wait for a checkpoint or a WU complete before switching apps.

---

BOINC WIKI

> Still, BOINC should wait for a checkpoint or a WU complete before switching
> apps.

Agreed. For one thing, doing so would negate the requirement of leaving apps resident when paused (e.g., CPDN). I lose up to ~20 of CPU time if I kill BOINC on a CPDN-enabled machine here. (Slow systems.) I'm looking forward to the new checkpoint/finished-WU logic.

---

> > Still, BOINC should wait for a checkpoint or a WU complete before
> switching
> > apps.
>
> Agreed. For one thing, doing so would negate the requirement of leaving apps
> resident when paused (e.g., CPDN). I lose up to ~20 of CPU time if I kill
> BOINC on a CPDN-enabled machine here. (Slow systems.) I'm looking forward to
> the new checkpoint/finished-WU logic.
>
One of the Alpha testers was losing 25 minutes per hour of CPDN crunch time with the remove from memory setting.

---

BOINC WIKI