Hi,
I've got an Alpha (Compaq) box with Linux for playing around.
For running squid in dedicated mode. Compiled with async-io.
I've got a problem that squid sometimes (quite often) gets
into 100% cpu mode without any traffic whatsoever. sometimes
one thread (linux process) sometimes all of them. box goes
crazy about it and quickly develops load of 20 and more.
I have no experience with neither Alpha severs nor Linux.
Perhaps you can suggest me things to watch and avoid or
do any specific way.
box itself is DS10 (EV6 at 466MHz) with 512M ram and
3x 9G scsi 10Krpm disks, 2 (16G) of which are for squid.
Linux is RH kernel 2.2.16
squid disks are ext2 with defaults and mounted noatime.
squid is 2.2S5-hno-20000202
compiled with compaq cc,
configured with:
CC=ccc ; CFLAGS="-O2 -arch ev6"
./configure --enable-dlmalloc --enable-icmp --enable-delay-pools \
--enable-htcp --enable-forw-via-db --enable-cache-digests \
--enable-err-language=Estonian --enable-underscores --enable-snmp \
--enable-async-io
no problems compiling, it runs, but sometimes hogs all the cpu.
I'm very unfamiliar with linux, I tried to strace squid, but it
seems that in pthread mode only main process can be traced.
Is this problem anything known? What should I look for first?
Before I'd dive into debugging with squid means, I'd like to ask
if alpha+linux is an ok selection, or should I rather try it with
digital unix, freebsd, or anything else?
Sort of question of which OS is best suited for Squid cache
running on Alphas?
And another problem. I need the box to be able to service upto
4000 concurrent sessions, and I stumped into FD_MAXSIZE problem.
Linux is configured for 8192 FD's per process, but it seems that
squid sources specifically disable FD > 1024 if it detects
Linux as OS. Is this because of some bad behaviour from Linux or
is it historical rudiment? The box I have supports poll(), threads,
and FD's > 1024 ok.
Received on Mon Jul 24 2000 - 09:07:28 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:54:35 MST