Hi,

The program developers very well know about the checkpointing and the regression to an early checkpoint, which particular on slower machines can lie apart quite a bit, but it's a trade off against writing very big files very frequently to disk.....see last paragraph below... both cases of a restart actually set them back to about 9%, where before BOiNC shutdown they were >51%.

The solution is to activate the hibernation feature. On restart, your work unit will take off exactly where it left off. Several helpful posts were made as how to set this up. My machine, sitting on UPS btw, goes in hibernation after 5 minutes or so and resumes right from the second it left off. Hibernation, opposed to Standby mode does not use a drop of juice.... it's a full memory state write to disk.

As to that penalization..... are u using the 5.7.x development version for this to happen i.e. the non-loss of CPU time. It's a balancing act i suppose and don't know if it will stay in it. Myself had 2 border line outliers, which in fact helped to pull up the median to above a credit of what would have been granted under the pre-outlier policy. Have already noted the observation today. Don't know if its an alpha BOiNC feature or a design into the science.

cheers

Okay great.... on hibernation there are good posts here and a revelation by LawrenceHardin, that if u dont see hibernation after setting it up, holding the shift key will make it show in the windows(?) closing menu.

For Genome Comparison and likely HPF2, when it restarts on BOiNC u will likely have to use at least 5.4 as it transmits using compression features.

Here's the vanilla download page: http://boinc.berkeley.edu/download_all.php

5.2.13 is geriatric / old if u will.

ciao

I have also come across this problem. I am using BOINC 5.4.11. Up until recently it could pick up its HDC work from where it had checkpointed, but now it will often start crunching the WU from the start. Maybe something wrong with the newer WUs...

It is essential to first snooze BOiNC. Close BOiNC and let time to write there is to write to disk, then shut down. Experimenting needs to be done whether the 'keep in memory' option in the prefs should be on or off..... not all PC's respond the same.

Also the latest HDC versions retain the CPU time spend till shutdown, even when loosing the progress due return to a previous checkpoint. This increases the claim, sometimes works out as becoming an outlier, sometimes helps to bring up the median to get more credit, opposed to having been run in 1 go.

Again, proper hibernation works. Standby will most likely not, as it is not writing all used RAM data parts to disk.

WCG/UD agent is pretty good at it too.

Yes, there used to be only 4 when HDC started, every 25% but the last series have distinct points of progress, seen 37.500, 62.500, 75.000 and 87.500. Not seen 12.500, but assume it's there.

As for the benchmark, that indeed happens, but only if the 'keep in memory' is not switched on. If switched on, there is no loss.

There are checkpoints whenever possible, it's though an elusive bug that during unidentified conflict moments cause progress loss.