On Wed, Aug 19, 2009 at 08:43:34AM +0200, Johan Duflost wrote: > Hello, > > Yes I know that session stickyness is enabled and this is what we need. > The problem is that when we make load tests with for example 10 concurrent > new users and 5 backend servers, we > noticed that some servers handle for example 4 users and other servers > handle only one. Is it normal?
no, but when running benchmarks, there is a well-known "resonance" problem which sometimes occurs. If your injection scenario contains a number of objects which is a multiple of the number of servers, you can sometimes end up with user sessions always created on the same server. You should check the stats and see how the load is spread in the LbTot column. If it is imbalanced between the servers, there is a problem (then check the Checks/Dwn column to see if one server is not flapping). Willy
