landolini,

It's not a good idea to power off frequently. You don't say how you do that' soft or hard, but yes there's always a chance, particularly when doing hard shutdowns.

With your frequent power offs, think you are better off with projects that have a short checkpoint interval such as HCMD2 and RICE. The HCMD2 on average is basically limited to 4 hours run time per job (can run longer if it has made it to 60% of all positions in a task), or RICE which has an absolute limit of 7 hours (plus a minute or so to finish the last seed). On top both these projects have very low memory needs i.e. what is saved to disk is very small, opposed to CEP which has very large work files.

edit: spelling

Officiously, best guess, you toyed with system dates and that caused BOINC to shift your BOINC Stats tab too. See the Start Here FAQ Index the preface item under section 4 before seeking any further help. When your client is a few months behind, the servers too wont allow buffering, so it becomes an operation of 'complete a job, return it, get a new one, then wait and see if the new one completes in time'. If years behind, the client will tell you to get a new certificate and no work at all is provided.

Think the latest client is more tolerant on this, but not going to test it.

In old , a few minutes is enough to upset allot of variables. Again, not so in the newer clients, but 10x per second reading and writing of client_state.xml does happen in circumstances. The housekeeping has changed to move more into the temporary job slots, away from the constant client_state.xml trans-actioning. Still not recommending 6.6 yet. Continued weirdness in 6.6.36 to include bald spot causing head scratchers and still not resolved in the 6.6.38 alpha.

AND yes, I'm sure there are thousands that are happy with 6.6.36 doing the job, but we like to prevent loosing thousands because things go haywire, out of control.

That "clutch" part as David Anderson qualified it himself, the rDCF, is not going to change anytime soon. A trac ticket 2 years old or longer.

As you will have read, BOINC was designed for 1 project 1 DC. WCG broke ground for multi-science-projecting and here we are several years later and still only 1 rDCF and not one by science/version. The common comment you get at the developers forum is something like "the projects need to do a better forecasting the job durations". It conveys to me complete ignorance when it comes to understanding basic concepts as none-deterministic calculations, currently shown through in the advanced job cutting algorithms the techs have developed here to keep the HCMD2 durations in check. One CMD2 job that has 1 position that takes 4 hours, not several minutes as the billion others do, and the client buffer is in bloat mode, all-projects across, speak the total of WCG.

Now, with more DC grids on multi science, the next coming Geneva outing could/should have put this on the agenda as a priority topic. knreed, please pass.

thanks.