S5R2c WU's
30 Apr 2007 17:52:13 UTC
Topic 192648
(moderation:
Four of my machines seem able to crunch the S5R2c WU's without the dreaded "Client error" code 11 SIGABRT.
Maybe all my other five machines have flakey memory, I don't know.
But now I am going to halt, for now, all but the four that seem to be hanging in there. I am wondering if this is being looked at as something in the Einstein app, or if this is just the way it is now.
When I get a chance I will run memtest on the machine here, but for now we're off to Shanghai.
...
S5R2c WU's
)
the current app appears to have some bugs in it, it's unlikely to be your hardware although why some machines are dieing constantly and others problem free isn't publicly known.
Bruce Allen has the same
)
Bruce Allen has the same problem (including some 4.18 results)
so I'm sure the project team will find out how to fix it.
As long as he has the same problems, you shouldn't try to fix
your machine as it is probably not broken.
No, it probably isn't. One of
)
No, it probably isn't. One of my hosts ran into "SIGABRT" with the 4.18 app yesterday and crashed a WU after 50 hours, but it has also returned some perfectly valid results (and shown no problems in memtest) so I'm quite sure the box is working fine. It is an AMD host, though... I've heard the opinion that AMDs are more likely to get these problems, just like they tend to perform poorer (although that mostly affects Windows, and mine is a Linux host... interesting...) with the current science app. So, don't worry about your boxes, I'm sure there's nothing wrong with them. Maybe they all got unlucky with the datafile and received more "difficult" WUs. The project staff is working on it...
RE: the current app appears
)
This problem is only affecting the Linux client, right? It is affecting AMD and Intel CPUs alike.
For my hosts I found that the version and/or presence of libGL (Open GL package) may also make a difference. After using LD_LIBRARY_PATH inside run_client to point to a fake libGL that doesn't load (after that the client disables grahics support but will carry on regardless), all my Linux boxes seem to do just fine.
Maybe someone with a box that is reliably failing on WUs wants to give it a try, if this can be confirmed we would be a big step ahead in resolving this problem.
Disclaimer: This will disable graphics for all of your BOINC clients, and some clients may even react adversely to this. So this is best tried out for a host that is running exclusively E@H and no other BOINC projects.
CU
BRM
I now see that with the 4.18
)
I now see that with the 4.18 version of the heirarchical search that things seem stable again. I also started up the 4.21 beta version on one machine and it produced two results with no problem.
When I see these problems I have an urge to debug, but in this case I have little clue what to debug.
...
RE: I now see that with the
)
4.18 was badly affected as well, but the beta (4.21) seems to fix this. Bernd said the beta will probably soon be released as regular download, so everything will be fine soon.
CU
BRM