My 1.1.20 on Solaris 2.6 x86 stalls

Hi folks,

I've been testing a new squid proxy for a few weeks and suspect I have a

Solaris/Squid/Config problem that I hope someone can shed some light on.

The setup is Solaris 2.6 x86 running on a dual Pentium Pro, 64MB
(->128MB real soon), 30GB RAID (of which 1GB is swap). The Solaris
install is vanilla with the recommended patch, no nfs, no NIS, no
Openwin/CDE, no routed. Squid is vanilla 1.1.20, configured as the sole
proxy for the site. A little more memory will help, but the machine
never rarely gets off idle with no more than 10 concurrent users. Most
of the time it works just fine, ...

.... but occasionally we see (IE3-4, Netscape 3-4) it stall for 5-30
seconds for no apparent reason. No interesting squid logs, no system
adm/messages, no syslogs, no console messages. When the problem occurs,
logged in shells running vmstat or top continue normally - showing
normal/no activity. The big clue (I hope) is that trying to open a new
telnet or rsh during the "stall" also fails to connect. They also appear

to be blocked until everything mysteriously comes good again.

Un-tuned (I'm building a task/test list of malloc, poll(), ndd,
/etc/system, sync, etc for later) I figured this configuration should be

good for a few hundred users without noticing a performance hit. We have

similarly configured machines running other apps that don't exhibit this

problem. I've been dredging the Sol and Squid FAQ's and list archives -
but nothing so far.

Any suggestions on where to look from here would be greatly appreciated.

(I'll summarise.)

Regards,

Bill Vickery
bill@arvida.net.au
Received on Tue May 19 1998 - 09:20:44 MDT