|
Hi Horms,
> If anyone is able to volunteer this information either
> on the list or to me privately then I would be most grateful.
> Please include some rough information on your system:
> What architecture, which kernel, which version of LVS,
> approximately how many nodes, and how busy it is.
We are using LVS-DR for 2 years now.
Setup is the following:
Large Website (no ssl)
1 director with lvs-1.0.7 and keepalived (running also in local node mode)
9 real servers, different hardware, therefore wlc scheduler (kernels from
2.4.18-2.4.20)
director is running 2.4.20, realservers different versions with hidden arp
patch
We have 2 VIPs, one for graphics and one for html.
The gIP has a peak of 500-700 active and 1500 inactive connections
the hIP has a peak of 20-40 active connections (disabled http-keepalive,
forcing http/1.0) and 4000 inactive connections.
We had 2 problems with LVS, but nothing directly connected to lvs stability:
1. I sometime managed to use different ipvsadm versions (one from debian and
one from source) and that gave unpredictable results. I had to cleanly setup
the kernel again. Much confusion, but no big deal.
2. When upgrading the realservers from debian potato to debian woody, we
sometimes had the problem, that the default route was not set, although it
has been entered correctly in the appropriate config files. The result was,
that _all_ connections went to the realserver without default route. Very
strange when you have 10 RIPs and only one gets all the traffic, choking
heavily under the load. Took some time till we realized the problem.
So the bottom line is that we never had stability issues that came from lvs
itself. Lvs works like a charm, at least for our type of use (plain http, no
persistence, wlc scheduler).
Regards,
Bjoern
|
