Re: [squid-users]Memory leak problem on epoll i/o squid on IA64

From: David Nicklay <dnicklay@dont-contact.us>
Date: Thu, 04 Sep 2003 11:33:42 -0400

Hi,

MUTHUKUMAR KANDASAMY wrote:
> Hello all ,
> In the compilation of squid ,I used cache_mem 1200 MB cache_dir null

The cache line looks fine, I doubt that is causing you problems

> fs.file-max = 16384

That should probably be higher, like 32786 or something higher...

> 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

What the heck? Why are your settings set to such insane levels? The
first number in net.ipv4.tcp_wmem and net.ipv4.tcp_rmem would cause you
problems.

tcp_wmem - vector of 3 INTEGERs: min, default, max
         min: Amount of memory reserved for send buffers for TCP socket.
         Each TCP socket has rights to use it due to fact of its birth.
         Default: 4K

tcp_rmem - vector of 3 INTEGERs: min, default, max
         min: Minimal size of receive buffer used by TCP sockets.
         It is guaranteed to each TCP socket, even under moderate memory
         pressure.
         Default: 8K

Taking a wild stab, and saying 256 file descriptors open, that means
something on the order of:

256 * (2000000 + 1000000) = 768 MB (and that is MINIMUM even under
moderate memory pressure)

Lower all of your numbers down to a sane level....

-- 
David Nicklay
Location: CNN Center - SE0811A
Office: 404-827-2698	Cell: 404-545-6218
Received on Thu Sep 04 2003 - 09:31:36 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:19:32 MST