How to solve mysterious system lockups?

[N. Harrington]

--- WorldLink at yahoo.com,
UNEXPECTED_DATA_AFTER_ADDRESS at .SYNTAX-ERROR. wrote:

> On Mon, 4 Jun 2007 19:46:30 -0700 (PDT)
> "N. Harrington" <drumslayer2 at yahoo.com> wrote:
> 
> > 
> > Hello
> >   I have several systems that are used as squid
> > caching servers. I have some systems that use SCSI
> > disks and some  that use SATA disks. They are
> > identical in everyway except for the sata vs SCSI
> > drives. 
> > 
> >  At random times, the sata based systems seem to
> be
> > freezing. You can ping them and they respond, but
> you
> > cannot log in. Nor are any logs processed during
> that
> > time. 
> > 
> >  I figure it mist be something to do with the
> disks,
> > but I am not sure how to solve it. There seems to
> be
> > little rhyme or reason. It does not happen
> necessarily
> > during busy times. It can happen in the middle of
> the
> > night.
> > 
> >  Any pointers in how to track down the cause would
> be
> > much appreciated.
> > 
> >  Tyan S2881 Motherboard - 4gigs mem
> >  Using 4 SATA (or scsi) drives
> >  FreeBSD amd64 6.2-STABLE.
> 
> Hi Nicole,
> 
> I was also having system lockup problems with my
> FreeBSD-6.x servers running Squid with SATA disks. I
> could ping it but could not login using SSH.
> 
> However, in my case, it was due to high mbufs usage.
> 
> What's your output of the following commands?
> 
> netstat -mb 
> sysctl kern.ipc.nmbclusters   
> 
> Thanking you...
> 
> > 
> >  Thanks!
> > 
> >   Nicole

   Hi
 I have my servers behind a load balancer that also
does TCP acceleration / offloading so my mbufs usage
is pretty low. (it manages about 24 connections tops)
 However I have been noticing a larger and larger use
of swap space. Either because of an increase on our
part in the number of files stored or from upgrading
to the newest squid versions (2.6-STABLE12 and 13)
 (last look showed me having 9 million objects in
storeage for that one server)
 
  So, we could be suffering from the same thing but
from different causes. I plan to be adding more memory
soon (from 4gigs to 6 and 8) as a test. 
 
 Also I have been adding SW_Watchdog to my kernels for
now. It seems very good so far at sensing the death
and rebooting. As well as soon adding some debugging
to it to hopefully capture something useful.


 Be well.

  Nicole