Clock Problems - possibly BOINC?

[Paratima]

I am running RNAWorld on two XP/SP3 boxes side-by-side. Both are ASUS motherboards with AMD 3 GHz X2 processors. Neither are overclocked. One keeps pretty good time all by itself, whereas the other gains maybe 6 minutes/day. Consequently, one relies heavily on system clock updates, either from Windows or another clock-setting utility.

The system with the regular clock has not had this problem. The fast box gets its tasks reset several times a day, like this:

8/16/2011 16:01:39 | | System clock was turned backwards; clearing timeouts
8/16/2011 16:01:41 | RNA World | Task cms_GA-p[N-O-Lin64f]_1_Nostoc-punctiforme-PCC-73102_CP001037.cir.EMBL_RF00350_snoZ152_1313176923_30475_3 exited with zero status but no 'finished' file
8/16/2011 16:01:41 | RNA World | If this happens repeatedly you may need to reset the project.
8/16/2011 16:01:41 | RNA World | Task cms_GA-p[PZ-Lin64f]_1_Photorhabdus-luminescens-subsp.-laumondii-TTO1_BX470251.cir.EMBL_RF00488_U1_yeast_1313370723_29239_0 exited with zero status but no 'finished' file
8/16/2011 16:01:41 | RNA World | If this happens repeatedly you may need to reset the project.
8/16/2011 16:01:41 | RNA World | Restarting task cms_GA-p[N-O-Lin64f]_1_Nostoc-punctiforme-PCC-73102_CP001037.cir.EMBL_RF00350_snoZ152_1313176923_30475_3 using cmsearch version 31
8/16/2011 16:01:41 | RNA World | Restarting task cms_GA-p[PZ-Lin64f]_1_Photorhabdus-luminescens-subsp.-laumondii-TTO1_BX470251.cir.EMBL_RF00488_U1_yeast_1313370723_29239_0 using cmsearch version 31

Needless to say, if there was a long-running XXL task being processed, this sort of thing gets old fast. When I turned off clock updates, the problem disappeared. However, I can not run that production system with an inaccurate clock.

The fast box is 11259; the normal one is 9903.

I just upgraded the fast box to BOINC 6.12.33 but it made no difference. Is this your problem or BOINC's?

[Ananas]

It's the old heartbeat bug in the BOINC API, it's somehow like this :

If the project client find an outdated timestamp of the core client in the shared memory, it assumes that the core client is dead and exits. This usually happens when

- the user (or some irregular PC behaviour) changes the PC time

- some program needs a lot of CPU and does not give enough time to the core client

- the core client is busy with a stuck communication

- the core client is busy with unpacking large compressed files, which are part of some workunit

The BOINC despot thinks about it as a great feature though

[Paratima]

Wonderful. Thank you.