Re: Kernel 2.2.19 crashes, 2.4.4 bad performance
Hi,
[.... jobs running into swap ? ...]
> True, but I've seen stranger things happen :-)
Me too ;-) But after checking my entries in /proc, the was no swap activity during
all of the benchmark runs.
[... CONFIG_RTC broken in 2.2.19/alpha SMP ...]
> I would recommend running an NTP daemon or something to sync the time on
> all of your servers on boot and periodically rather than relying on the
> RTC. It tends to be more reliable when you're dealing with multiple
> servers anyway (rather than having to set time that's fairly close on
> multiple machines by hand).
I do already run ntp synchronized to an external clock-source. But until ntp gets
started, various files have their atime and ctime already set to a date in the year 2049.
That, of course, sucks. Also, in my ntp-setup, a clock-skew of that difference, causes
ntp to write to logs instead of happily adjusting date+time back 50 years ;-)
I could configure it otherwise, of course, but I'd rather stick with my current setup, whch
is necessary for various reasons.
Set aside the fact, that i strongly dislike working around obvious kernel-bugs that way.
Currently, I'm poking around the rtc-driver in an attempt to fix the stuff myself.
Therefore the question, if any patch is already available ( maybe in 2.2.20-pre.. ) ?
[.... on the new aic .vs. old aic driver ...]
> I haven't looked at it too hard yet. I personally like the old driver
> better, but that's only because I'm more familiar with it code-wise :-P
Well, the new driver did not show any obvious problems yet. Since my setup's purpose is
(partly ) to verify configs, that are duplicated in customers machines, I maintain, my
personal preference has to be to stick to mainstream configs as close as possible - on
*that* machine, that is ;-)
[ at this point, it should be mentioned, what a *great distribution* Debian is, when it
comes to running business critical/relevant jobs. Thanx and respect to all involved.
This has to be said, once in a while, IMHO ;-) ]
> Was there much net traffic when you ran the tests or was the 3Com mostly
> inactive?
I'd say, mostly inactive. However, the UP2000 runs headless with an x86 serving as
multihead-Xterminal. So all traffic between Xserver on the x86 and xterms, fvwm2 and
the like on the alpha goes via net. So, there is *some* traffic, but i'd hardly call it
significant.
> Not yet. I'll see if I can free up some time to look at it more
> in-depth. Sounds like you definitely have a strange case,
> though. Perhaps I'll run some tests on my UP2k at work over the coming
> weekend and see if I get similar results.
Mee too. I currently pushed the issue a bit down on my to-do list and reverted back
to 2.4.4. But i'll look into it in the next days and provide further details.
Regards,
T. Weyergraf
--
Thomas Weyergraf kirk@colinet.de
My Favorite IA64 Opcode-guess ( see arch/ia64/lib/memset.S )
"br.ret.spnt.few" - got back from getting beer, did not spend a lot.
Reply to: