IMHO this needs to be handled by the project, submitting big jobs only to PCs which run 24/7 and/or have leave-in-mem=yes.

IMHO this needs to be handled by the project, submitting big jobs only to PCs which run 24/7 and/or have leave-in-mem=yes.

Which will require a modification to Boinc to pass that information back to the Rosetta servers.

17128664 seemed to be going wrong. It was still under 2% finished after more then an hour of crunching. It was moving, because I'd noticed it at 1.87% and later saw it had crawled to 1.9%, so not a 1% issue.

Thing is, next time I looked at it, ~5 minutes later, it was complate and ready to report. Note despite a 7200 seconds preference, it ran for a lot less.

17128664 seemed to be going wrong. It was still under 2% finished after more then an hour of crunching. It was moving, because I'd noticed it at 1.87% and later saw it had crawled to 1.9%, so not a 1% issue.

Thing is, next time I looked at it, ~5 minutes later, it was complate and ready to report. Note despite a 7200 seconds preference, it ran for a lot less.

The WU will start at 1%. Then there will be small increments as the WU goes through several stages of the first model (less than 1% total). After each model a larger jump in percentage is made based on how many models the rosetta app thinks it can do. If it doesn't think it can fit another model in without running too far over, the percent will jump directly to 100%.

The "leave in memory" is no longer an issue, but Rosetta "check-points" at the end of each model (notice the Model/Step info in the screensaver). AMD_is_logical explained it very well.

Most WUs are small proteins, which take only ~10min per model.

Some recent ones are very big which take 2-4hr per model (on Pentium4!). So to finish such a WU, Rosetta needs to run on your PC for 4hrs, WITHOUT being unloaded from memory. If a PC unloads Rosetta every hour to run another project, it will never finish, as it'll start everytime from scratch.

The surest way to run Big WUs would be check the "leave in memory when preempted"=YES.

IMHO this needs to be handled by the project, submitting big jobs only to PCs which run 24/7 and/or have leave-in-mem=yes.

The WU will start at 1%. Then there will be small increments as the WU goes through several stages of the first model (less than 1% total). After each model a larger jump in percentage is made based on how many models the rosetta app thinks it can do. If it doesn't think it can fit another model in without running too far over, the percent will jump directly to 100%.

If this was the case, would not the details of my results show that only 1 structure had been produced? This is not as observed in the results for that machine.

...
If you have a largescale WU which would take about 3 hrs for one model and your run time is set at 2 hrs, then the WU is gonna complete that model.

Even if it's 1 hr above your settings.

And the progress procentage will start at 1% increase in small steps towards 2% all the way up to 3 hrs CPU time and then jump to 100% and completion.

- Knorr

[b]"I'm trying to maintain a shred of dignity in this world." - Me[/b]

Having "noticed" this behaviour, I have taken to watching that machine more carefully. I can now report that all the wu's I've watched have done the same thing. The wu progresses very slowly up to around 2% then finishes. This is not the same as wu's on this machine.

Here, my current wu has 1:04:33 and 37.4% done, the box in question has 1:59:46 and 1.52% done. Both machines have work unit length set to 7200 seconds, as can be seen in the results.

The *_largescale_large_fullatom_relax_* WUs are very big and sometimes "Steps" remains at 0. Usually just one "Model" will fit in the 7200 seconds (2hr) timeframe, in which case the %-progress indicator may stay at e.g. 1.5% for 1-3 hours while computing the first model and then finish, realise that it can't run a second model per your WU-runtime settings (7200 sec might have been already exceeded), so it jumps to 100% and finishes.

I use 8-hr WU-runtimes (instead of 2hr default) and a big WU taking 2hr per Model might jump 0% -> 25% -> 50% -> 75% -> 100% in BOINC progress.

Certainly, the wu's I have running now should be with the new target time, and they are presenting VERY differently on screen as I look. Both are running largescale_large_fullatom wu's, on one machine it shows 01:11:44 and 1.58%, the other, 1:14:06 and 27.27%.