There's as eluded 16 checkpoints in a CEP2. On my 8 core I7 lappie, the first and second are within several to 10-15 minutes, then the 3rd job (#2), takes 3-5 hours. BOINCTasks (a 3rd party BOINC Manager) will actually count down the checkpoints, but you can see in the stderr.txt file of the job slot, typically on W7 in C:\ProgramData\BOINC\Slots\#, where the progress of a task really is. The stderr.txt file is the same as what's uploaded as Result Log which you can visit on the Result Status page. A sample on a handstarted CEP2 task, now working on the 3rd job, is this:

INFO: No state to restore. Start from the beginning.
[10:18:23] Number of jobs = 16
[10:18:23] Starting job 0,CPU time has been restored to 0.000000.
[10:23:12] Finished Job #0
[10:23:12] Starting job 1,CPU time has been restored to 277.993782.
[10:38:10] Finished Job #1
[10:38:10] Starting job 2,CPU time has been restored to 1144.672938.

(First job is numbered 0, and any job takes the output from the previous job as input)

Note that a CEP2 task assumes 12 hours run time on systems that never finish them completely [12 hours is the cut-off]. The progress percent is then based on the 12 hours assumption.

Setting write interval has no effect on CEP2... 2 minutes or 2 days. It writes the output to disk only at end of a job, else way too big, in the gigabytes size which would be disastrous if you were using the computer as heavy disk IO really impairs the user.

If a whole job takes 7 hours, and there are no write points but the once observed, the question is, what then is logged in the stderr.txt file after long run time and prior to hibernating? When resumed from hibernating, there should not be an interruption logged in the the stderr.txt file. If there is, what does it look like? When restarted, the log shows from which point and what seconds CPU time that resume is taking place.

Of old, shutting down too fast, not hibernating or standby, could cause damage to a task, hence when this is observed, stopping BOINC first, waiting a minute, to let the writes to disk complete, then shutting down would prevent that. Typically this happened on Vista. It was fixed with later BOINC clients.

I have tested on two CEP WUs to stop boinc manager before stopping win 7 now:
That works. I recognised that when CEP WUs are running on boinc manager the manager needs approx. 10-15 sec to shut down. Obversly windows is not giving boinc manager that time when win itself wants to go off.

Thanks for your help

I think I have this kind of problem because WIN is running on a SSD, so very fast, and boinc "only" on a normal SCADA HDD.

To add, an interesting key found in the same place as WaitToKillServiceTimeout, is PreshutdownOrder. A google popped out this MS support article: http://support.microsoft.com/kb/146092 (but no word of PreshutdownOrder) and this http://serverfault.com/questions/34427/windows-service-dependencies (for the adventurous and inquisitive)

edit: more expansion on PreshutdownOrder in this article that is linked in the one above: http://blogs.technet.com/b/askperf/archive/20...n-and-crash-handling.aspx

I loaded some CEP2's on my wife's laptop. By accident, I think, because that machine is shutdown every evening and sometimes even down for more than a day.
Because of the known long intervals between the checkpoints (sometimes even 6 hours), I decided to test the sleep and hibernate modus for this in stead of letting the laptop run for the night.

Both modi were working fine, but before switching to sleep or hibernate, I suspended all tasks first with 'leave application in memory on'.

After restart I resumed the CEP2-task and it restarted exactly from the point where I left it without any loss.