Simon Matter wrote:
>
> Hi Bret
>
> Bret Hughes schrieb:
> >
> > I have a configuration on 6 boxes, Duron 700 with 128 MB ram and 10GB
> > ide harddrives. These machines are all XFS 1.0.1 using the kernel from
> > the 1.0.1 RH install iso and most of the RH updates.
> >
> > They were all placed into service the same day and yesterday 5 days
> > after they were turned on they all stopped responding to ssh. A couple
> > would still respond to a ping but nothing else. Actually I caough one
> > before it quit altogether and rebooted it last night. These machines a
> > kiosk type display that scroll html and flash pages using (netscape as
> > the browser). There is no user interaction infact not any input device
> > at all. A different page is displayed every 10 seconds.
> >
> > Now, After rebooting them I can get to the sadc data and looking through
> > the logs shows really bad stuff happening at 1:00PM yesterday on the one
> > machine I have really looked at closely. interrupt 14s out the wazoo
> > and what I think must be the cause, the dentunusd goes to 0. Looking
> > over the logs of the last few days I can see the dentunusd creeping down
> > from the beginning at boot of over 20K to 0.
> >
> > Since netscape is such a pig I kill it and restart it once an hour and
> > restart X once a day. both these events seem to take a tremendous toll
> > on this value and never gain it all back.
> >
> > I don't know if this is an XFS issue or not but I thought that I would
> > start here. Any ideas? I can send all the log data anyone might use if
> > it would help. Right now I am going to reboot the damn things nightly
> > like they were windows machines for Christ's sake.
> >
> > BTW the same scenario and control scripts run for months on end on RH
> > 6.2 using the 2.2.3-16 kernel.
> >
> > Any tips and or other places to ask are appreciated.
> >
> > TIA
> >
> > Bret
>
> Did you run the RH-6.2 stuff on exactly the same hardware (including
> same disks?)
No, I have not. The load is really not very high though. I will go
ahead and try the 6.2 config on the new hardware. Something is
definitely weird.
>
> What about the disks, are they running with DMA enabled?
>
Yes. the default was to enable mda and the 32 bit access. I have not
increased the bus speed though I plan on testing it.
> Over here our mailservers used to be IBM PC's with Fujitsu Harddrives.
> After some time of operation they started to slow down and load started
> to increase. They even did not respond to ssh sometimes. In the end I
> saw that those disks failed under the heavy load of the mailserver and I
> replaced them with other disks. We have hundreds of the same disks in
> windows desktop PC's with no problem. Linux just pushes the hardware
> more.
>
> Simon
Thanks for the thoughts.
Bret
|