[Vista] Avoiding client error exit code 1073807364 (0x40010004)
10 Sep 2007 0:16:16 UTC
Topic 193121
(moderation:
It appears that this error occurs when Vista shuts down before either BOINC or the Einstein app has time to close properly.
At least two ways to avoid this:
1.) "Exit" BOINC Manager, not just close (as reported previouly for another exit error code).
2.) Chose option "suspend when user is active" and not to leave the Einstein app in memory.
- exit code 1073807364 (0x40010004)
2007-09-04 00:15:47.6091 [normal]: Start of BOINC application 'projects/einstein.phys.uwm.edu/einstein_S5R2_4.30_windows_intelx86.exe'.
2007-09-04 00:15:48.1863 [debug]: Reading SFTs and setting up stacks ... done
2007-09-04 00:16:05.6115 [normal]: ERROR: Couldn't open existing checkpointing toplist file
2007-09-04 00:16:05.6115 [debug]: Couldn't open checkpoint (2) - starting from beginning
[Vista] Avoiding client error exit code 1073807364 (0x40010004)
)
There's also the registry hack. That will slow down Vista's shutdown.
Kathryn :o)
Einstein@Home Moderator
I know there is a registry
)
I know there is a registry hack, and I have done that, and it still seems to only work sometimes.
Since they have been sending me these huge jobs that take over 40 hours of processing time, I think only about two of them have successfully gone through to completion on my Vista Home Premium system.
If I were working on a project that would take 40 hours, and didn't save my work at least once per hour so that I could go back to the previous version if anything went wrong, anyone would say I was silly, and have little sympathy for the fact that it resulted in losing my work and having to start again.
These tasks are supposed to be background tasks which should be able to just run through without user intervention from system users.
I think that part of the logic of such tasks should be to regularly checkpoint their progress, at least (say) once per hour, and in the event of a failure they should be able to resume from the previous checkpoint. Possibly the whole task could be regarded as failing if it has to re-start from the same checkpoint more than once or twice.
If there is a possibility of the checkpoint files being left in an inconsistent state when interrupted during a shutdown, the checkpoint process should keep an intact backup of the previous checkpoint information for recovery.
I gather that this problem seems to affect ALL Vista users, so it will become more and more of a problem to the whole BOINC system as a larger and larger proportion of newer faster computers are running Vista as time draws on.
Oh and by the way, the "registry hack" possibly should be automatically applied by the BOINC installer, and/or administerable through the next version of the BOINC manager to make it transparent to end users.
Would any of these measures be too hard to implement to be worthwhile? I suspect that if not addressed the whole BOINC system (or at least the Einstein project) would gradually grind to a halt month by month.
RE: Would any of these
)
The Einstein application already checkpoints. My latest couple of many from my log (with the checkpoint_debug flag on in cc_config.xml):
04-Oct-07 3:03:56|Einstein@Home|[checkpoint_debug] result h1_0520.20_S5R2__162_S5R3a_0 checkpointed
04-Oct-07 3:06:24|Einstein@Home|[checkpoint_debug] result h1_0520.20_S5R2__162_S5R3a_0 checkpointed
04-Oct-07 3:09:02|Einstein@Home|[checkpoint_debug] result h1_0520.20_S5R2__162_S5R3a_0 checkpointed
The Einstein application also makes a backup of the last checkpoint it was at. See this post by Bernd Machenschalk, one of the Einstein developers.
As for the registry hack to be implemented in the installer, I don't think so. BOINC 6.0 will have a totally different approach on its installation, mostly as a service only, running as a new user. You can read all about it here.
Also the new Windows beta app
)
Also the new Windows beta app aims at testing a feature that will make checkpointing more robust. I certainly think that BOINC should NOT change system-wide registry settings on MY machine.
CU
Bikeman
RE: As for the registry
)
When are we likely to get BOINC 6.0?
RE: When are we likely to
)
It hasn't even been released to the alpha testers yet. I wouldn't look for a stable version for quite a while.
As a comparison, the first 5.9 (development line for 5.10) was released around the beginning of April and the first release of 5.10 was around the end of June. So three month.
See the change log.
Kathryn :o)
Einstein@Home Moderator
Stealing this entry from the
)
Stealing this entry from the BOINC forums and putting it here, Nicolas won't mind I think.
I have had 5 client error on
)
I have had 5 client error on vista pinned it down to not exiting before updateing or down loading new programs that need a restart 1 odball 1 in progress 1 on stand bye the 1 on stand bye was the error with 000 sec ooo credit one in progress no problem
Peter, you bring up several
)
Peter, you bring up several great observations. Even though my Vista system has failed to complete both Einstein and Climate Prediction projects and seem to have a varity of exit codes, they all seem to have something to do with an inability to open or find a key file. Could this be something with disk caching services in the Vista environment? Is Vista shutting down the computer before the disk cache has a chance to empty? Even if this is true, my general preferances tell the applications to write to disk every 45 seconds so the most it should lose is that interval of computing. Maybe the fix is to get these projects to write-through to disk more often. This also seems to be confined to just these projects as Seti and Rosetta are unaffected.
As Peter observed, these projects should run with little user interferance in the background. Myself, I am at wits end. There is something very wrong when a project can get to 70 - 90+% of completion and not be able to find or open a key file as several of mine have done. I made the registry change, I don't leave applications in memory, decreased the write times, I even went as far as scrapping the entire Boinc directory and starting fresh to no avail. What else can be done?
RE: Even if this is true,
)
Checkpoints are written by the science application, not by BOINC. So even though you set a write to disk interval of 45 seconds, doesn't mean that either Einstein or CPDN write to disk every 45 seconds. Furthermore, it's a suggested interval at which time BOINC will write to disk. It doesn't have to be followed.
You did install BOINC to a directory of its own outside the C:\\Program Files directory? If you did, then you may want to try BOINC 5.10.42. Since 5.10.35 a lot of fixes have gone into the 5.10 engine to accommodate Vista. Not everything is fixed yet, for that you have to wait until BOINC 6 is released. Which can still take a while.