[squid-users] Memory leak on epoll support in IA64

From: MUTHUKUMAR KANDASAMY <kmuthu_gct@dont-contact.us>
Date: Thu, 4 Sep 2003 11:32:40 +0530

Hello Adam and all,

I am muthukumar working in epoll development on Squid.I am working in IA64
platfrom.Then I have tested the squid with epoll for 300 req/sec.But after
that it is consuming more than 1.9 GB out of 2.0 GB.As Adam's advice, I
have changed the swap memory from 1945.634 to 256.000.

[Try shrinking your swap partition in half (to about 1 GB) and see if that
improves your memory situation. You don't have to reboot the
system - just use swapoff to turn off the swap space, use parted to resize
it (check /etc/fstab for the minor partition # first), then use
swapon to turn it back on ]

But still the squid is exiting due to
Sep 4 11:10:19 pandia squid[1184]: Squid Parent: child process 1186 started
Sep 4 11:15:40 pandia kernel: Out of Memory: Killed process 1186 (squid).
Sep 4 11:15:40 pandia squid[1184]: Squid Parent: child process 1186 exited
due to signal 9

I have tuned the kernel parameters of
net.ipv4.ip_forward = 1
net.ipv4.conf.default.rp_filter = 1
kernel.sysrq = 0
kernel.core_uses_pid = 1
fs.file-max = 16384
net.ipv4.ipfrag_low_thresh = 900000
net.ipv4.ipfrag_high_thresh = 1000000
net.ipv4.ipfrag_time = 45
net.ipv4.tcp_rmem = 2000000 2250000 2500000
net.ipv4.tcp_wmem = 1000000 1250000 1500000
net.ipv4.neigh.default.gc_thresh1 = 1024
net.ipv4.neigh.default.gc_thresh2 = 4096
net.ipv4.neigh.default.gc_thresh3 = 8192
net.core.rmem_max = 1500000
net.core.rmem_default = 1500000
net.core.wmem_max = 1000000
net.core.wmem_default = 1000000
I have used the squid configuration of
configure options:
============
'--prefix=/usr/local/squidepoll' '--enable-epoll' '--disable-poll'
'--disable-select' '--disable-kqueue' '--enable-storeio=null,aufs,ufs'
'--with-file-descriptors=16384' '--enable-async-io=16' '--with-pthreads'

I have tried a lot to satisfy more than 300 requests.But i cann't make it
.Any other parameter must be tuned for good performance?.

So please suggest me some information regarding to this.Then regarding to
kernel I am using the sys_epoll enabled 2.4.20 on Ia64.I have made it with
sys_epoll patch and IA64 patch.Using the sys_call in entry.S,i have used the
epoll support on IA64.

I want to know whether this affection is on IA64 kernel or
squid-3.0-pre3.Regarding to the 300 req test also consumed upto 1.8 GB.So
please tell some way to improve the squid for epoll support.The kernel
series 2.5 and 2.6 is supporting sys_epoll.But is their any affection of
kenrel on Squid-3.0-pre.

I have enabled the epoll on 2.4.20 mannually without default on 2.5 and
2.6.So I think ,there is no effection of kernel on Squid-3.0 in the effect
of memory leak.
Anyway to improve the squid-3.0-pre3 for epoll support.

Thanks in advance.

<Muthukumar>
Received on Thu Sep 04 2003 - 00:13:45 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:19:31 MST