[lvs-users] Help with diagnosing performance issue 60K connections graph

To: " users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: [lvs-users] Help with diagnosing performance issue 60K connections graphs provided
From: Malcolm Turnbull <malcolm@xxxxxxxxxxxxxxxx>
Date: Thu, 12 Mar 2009 09:18:02 +0000
Hi all,

(sorry last post was in html by mistake)

I've been scratching my head a bit with a performance issue on a live
system and was wondering if anyone could point me in the right
direction for resolving?

I have a load balancer using 2.4.32 stock kernel running LVS with HTTP
& HTTPS in DR mode with 10 real servers each.
The load balancer is a 2.4GHz Celeron with 512Mb RAM, broadcom gigabit nics.

free -m shows 300MB used + 200MB cached (no swap file)

When the active connections climb to 60,000 and throughput is about
100MB+ we get a sudden and massive drop of in throughput...
We get a corresponding drop in system CPU and increase in Kernel CPU...
I have a selection of graphs here including throughput and cpu usage etc.

No errors in syslog relating to NICs or netfilter etc.

The actual number of connections is actually about 9K or 11K/sec
(Direct routing exagerating them), I was looking at using ipvsadm
--set to reduce the TCP timeout and therefore connection table size
but I don't think that the memory is an issue.

Any other ideas?

NB. I've tested similar hardware to 800MB+ and 50K cons/sec (obviously
with small/large packet sizes though not live data)

Many thanks in advance for any insight :-).


Malcolm Turnbull. Ltd.
Phone: +44 (0)870 443 8779

Please read the documentation before posting - it's available at: mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to

<Prev in Thread] Current Thread [Next in Thread>