Manuel A. Matias wrote:
>
> Dear All,
>
> I would like to present you what I think is a quite strange behavior
> of a Dual Compaq machine (model AP 400) running Linux Mandrake 6.1, whose
> main use is numerical computation.
>
> The Distribution correctly chooses an SMP Kernel, namely the
> 2.2.pre13-7.
> Initially the system (that has 2 Pentium II at 450 MHz) runs
> correctly (each CPU goes faster against a test machine that has
> Linux running on a single Pentium II at 400 MHz), but after 2
> days, more or less, strange things happen, and the dual (still
> 1 job for each CPU) runs something like 4-5 times slower than
> the test machine, while it gets completely confused regarding dates.
> E.g, while making "ps -aux" the process corresponding to this
> command appears to have been launched a few days ago.
> Instead, the command "date" gives the correct date.
OK, the bogus times from "ps aux" sounds like the problem I had on RH6.0
(fixed on RH6.1), where I had to patch out a bug in libproc.so (from the
procps RPM). Basically, it tries (and on SMP it fails) to figure out
the value of HZ dynamically with a poor algorithm. That's not a serious
problem though, and the only symptom I ever saw was the rubbish output
from "ps auxOT". Perhaps you should check for other binaries which use
libproc.so?
I can't answer the problem about your slow-down though. Is it slowing
down as reported by the system or in real wall-clock time (and thus
obvious to a human)?
> I want to stress that both things (processes go slower and
> time gets confused at some level) happen at the same time.
> After it is booted things go correctly again, but only for a while.
I suspect that the "ps axu" problem is there from bootup but only
becomes really noticeable a good few hours after bootup when things
start to look really silly in the 'ps' output. It's not unprecedented
for me to be wrong though ;-))
Neil
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/dmentre/smp-howto/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
