"Brian J. Beesley" <[EMAIL PROTECTED]> writes:
> Have you run P-1 on an exponent around this size before? Just a
> thought - you may be running out of memory when stage 2
> starts. Alternatively I think with Fedora you are probably running a
> 2.6 series kernel - still experimental so far as most people are
> concerned; there may still be bugs in this, in any case there are a
> _lot_ of kernel tweaks going on in 2.6 & it may be worthwhile
> upgrading to the latest kernel. Fedora may well have a more recent
> kernel package if you can't be bothered to compile your own.
It is the RedHat kernel, not 2.6. It doesn't work on RHEL 3.0 either,
which has a heavily patched 2.4 kernel with some features from 2.6.
The RHEL kernels, which probably have a lot of patches in common with
the Fedora kernels, is rotten in a lot of ways. The system becomes
totally unresponsive swapping or other during heavy I/O, and mprime
can't do stage 2 of P-1.
I have also seen errors of this kind from more than 10 otherwise
healthy machines with RedHat kernels:
ERROR: SUM(INPUTS) != SUM(OUTPUTS)
I think this must be due to a driver not restoring FP register state
properly, because I can't reproduce the errors with a stress test when
the machine is otherwise idle.
I have been running a vanilla 2.6.7 (with some og my own patches) on
my RHEL WS 3.0 machine with an uptime of 115 days now. Which is
approximately the number of days since 2.6.7 was released. Never a
segfault in stage 2 of P-1 factoring and mprime, and none of those
errors.
Btw, I have found two machines where stage 2 mysteriously works with
RHEL WS 3.0 and RedHat kernel. The machines are otherwise identical
to other machines where mprime segfaults on stage 2. One Dell 2650
with Xeon CPU and one "no brand" with a PIII, both with 2GB RAM.
--
Sturle We know that dictators are quick to choose aggression,
~~~~~~ while free nations strive to resolve differences in peace
-- George W. Bush
_______________________________________________
Prime mailing list
[EMAIL PROTECTED]
http://hogranch.com/mailman/listinfo/prime
