BOINC, Einstein and the weird SIGFPE.
24 Dec 2007 22:06:48 UTC
Topic 193400
(moderation:
Hello everyone!
As you can see here Einstein has quite some problem on both of my PCs.
Most of the Workunits quit with Signal 8 (meaning SIGFPE.)
I was not able to really track down what happens, but it happens for some time now...
I even changed to the latest BOINC version without any success. :(
I'm using Linux and heard, that your Linux client often has some problems, so... I guess this is a bug that should be fixed or something like this.
Anyways, the problem is that I'm doing work for Einstein for MONTH now with just a few valid WUs and that is driving me nuts... it's just wasted CPU power, as well as for you, as for "me". :(
So do you have any clue how that happens/how it could be fixed?
-Ionic
BOINC, Einstein and the weird SIGFPE.
)
What distribution do you use? I'm sure this is not an Einstein bug, looks more your configuration or hardware is broken.
I'm using Gentoo GNU/Linux,
)
I'm using Gentoo GNU/Linux, but I don't think that my hardware or configuration is broken because Einstein is the only program that has such problems.
Even SETI is working really fine...
I've run memtest86+ for almost 48 hours when I got the PC - with any bad results.
Neither do I "rice", my CFLAGS were -O2 -march=pentium3 -mtune=pentium3 -mcpu=pentium3 -pipe -fomit-frame-pointer on one PC and -O2 -march=pentium-m -pipe -fomit-frame-pointer on the other one.
It seems much like an Einstein client issue, but I might be wrong as well.
-Ionic
Hi! The quoted error is
)
Hi!
The quoted error is quite rare and it's strange that you have two PCs consistently suffereing from them. Are your PCs overclocked? are they co-loated in the same room? You are not living in the vicinity of a nuclear reactor, right :-) :-) (just kidding)?
For the Pentium M, you might want to try the Linux / Intel "Power User App" discussed here. Unfortunaletly your PIII-S doesn't support it.
This version does floating point calculations differently. Whatever is causing this errors, with some luck it won't occur in the Power User app.
CU
Bikeman
BTW: The fact that other programs are not crashing is not really proving anything. The E@H app contains special code to catch these error conditions, older versions of the E@H would have continued as well for some time unless it ran into some random secondary error condition.
Hey, no, I would never
)
Hey,
no, I would never overclock any PC. :)
Heh, yeah, they are on the same room, but that room is quite cool everytime. I don't like hot temperatures. ;)
Thanks, I will have a look at that...
One more thing: I guess the Pentium-III-S's and Pentium-M's architectures are quite close together, isn't it? Well anyways... I will try to upgrade to the latest glibc in a few weeks (I'm using 2.5 whereas 2.6 is current stable) and rebuild everything, in the hope, that it might magically solve some of my problems. :)
And I will try that other binary on my Laptop, I hope it'll solve at least the one on my Lap. :)
Thank you for your help,
-Ionic
RE: One more thing: I
)
Indeed, the Pentium M is an enhanced Pentium III-S. Among other things, support for the SSE2 instruction set was added and this is required by the current build of the Linux/Intel Power user app.
CU
Bikeman
Yes, I know that too well.
)
Yes, I know that too well. :)
I made a full backup of my Laptop before getting it repaired, mounted the harddisk via loop on my Desktop and nearly EVERY command produced SIGILL - that was my first contact with SSE2 No-SSE2. :)
Anyways, I hope I'll have some luck with that.
SIGFPE is indeed rather unusual, on my Router (200MHz MIPS32) I could "produce" it by installing binutils-2.18 with uClibc 0.9.27... it was horrible, every binutils program exited with SIGFPE and did just nothing. Later, I read that binutils-2.18 is not compatible with that old uClibc anymore... bad luck. :/
Thus I thought that my "old" glibc could cause that problem and I hope that upgrading will help... we'll see. :)
By the way, Dual PIII-S rocks, isn't it? Unfortunately, I just have a single board, but two of those CPUs. Though, buying a dual board would cause astronomic costs because I would have to replace nearly everything, even the RAM (EEC is a must on dual boards). :(
Hope you enjoy it anyways. :)
-Ionic
Hi, well you are indeed
)
Hi,
well you are indeed right, with that new version I do not get SIGFPE anymore, but this time some other errors which look just MORE critical, could you please have a look at it? (The last result on sui.lan.)
-Ionic
Do you have installed some
)
Do you have installed some unstable libs?
Not that I would know.
)
Not that I would know. :(
-Ionic
Or made a major GCC upgrade
)
Or made a major GCC upgrade without rebuilding system libs?