My 9x host is having some problems as noted in the title.
ABP is faulting out at startup with an illegal instruction error.
GCE starts up OK, but aborts shortly thereafter with an out of memory error. This was noted with the 03 app version, and I saw the the post from Bernd about the 04 version fixing a memory leak so this problem may already be fixed.
It should also be noted that this host is one of my AMD old timers which is pretty thin on RAM by current standards, but all of that is known to the scheduler ahead of time and should be taken into account by the scheduler before being assigned tasks. ;-)
It turns out the illegal instruction problem for ABP isn't limited to 9x. This task faulted out on an AMD old timer running 2K Pro, and this one on 2K Server.
Alinator
Copyright © 2024 Einstein@Home. All rights reserved.
Problems with ABP and GCE on 9x
)
Hi!
I'm afraid the ABP2 app requires at least SSE support, and if your oldtimers don't have this, they will not be able to run this app.
The memory leak you mentioned was very, very small, I doubt that fixing this leak will free enough memory on these oldtimers. Sorry.
Bikeman
RE: Hi! I'm afraid the
)
Yep, it looks like the problem for ABP is the lack of SSE support in K6-x processors. It would be interesting to see if a PIII+ or Athlon XP+ running 9x would handle them. My PIII host is dual boot (9x-2K Pro), so I'll probably try that out when it draws its next ABP task.
For GCE though, as you suspected the pretty puny memory leak eliminated with the current .04 app wasn't the cause of the abort. It looks more like the problem is with 9x memory management. I have a GCE task running on another K6-x host (2K Pro) which is RAM limited to 192 MB (motherboard restriction) without issues (virtual twin to the one which aborted).
I can test this out with the PIII as soon as it finishes it's current task, since a GCE is the next one up in the bullpen. If it aborts on it running 9x, then the reason is most likely the app was was compiled with a tool which doesn't fully support 9x anymore (like VS 2008 for example). That might be able to be fixed by recompiling with VS 2005.
In any event, this all indicates the scheduler should be taking a little bit closer look at the hosts' capabilties and characteristics before it decides what it's going to send out for work. ;-)
Alinator
OK, K6-x Host (2k Pro) 251595
)
OK, K6-x Host (2k Pro) 251595 just pulled an ABP task to refill its cache with the approaching completion of its GCE task.
However, it will most likely be around 24 to 36 hours or so before EAH comes up in the run rotation again, and probably faults it out.
When that happens it should pretty confirm that the scheduler needs to be adjusted to exclude anything older than a PIII or Athlon XP from being sent ABP.
Alinator
Update: 251595 running 2K
)
Update:
251595 running 2K Pro completed its GCE task successfully and it validated.
This tends to support my hypothesis about GCE when it tries to run on 9x. I just noticed my PIII drew a GCE yesterday afternoon, so I'll boot into 9x and try to run it. That should confirm the problem isn't a hardware related one if it aborts as well.
Alinator