IHS / WebSeal - strange load-balancing problem.

[Pete Bennett]

We are seeing a strange situation in a performance test, where if we send 60
user load down through one HTTP server / WebSeal combination (on Win 2003)
to WebSphere 6 application servers, we get average response times in the
order of 3 to 4 seconds.

If we balance the load (from Network Despatcher) to two identical HTTP
server / WebSeal servers, and ramp up the number of users to 120, we get
average response times of around 1.4 seconds.

We have a load/performance test environment, which comprises Compuware
QALoad players out at a remote site going through Edge servers, then to HTTP
server / WebSeal machines. These machines have various junction points,
including redirection to an IIS server which is just a document image
store/server and of course, direction to the WebSphere 6.0 application
server machines, which are Sun Solaris.

None of the machines, nor the (Gb) network seem heavily loaded, with the
HTTP servers ( 4 x Xeon processors ) running at a maximum of 34% CPU and
about 2% network.

I just cannot imagine why the 60 workload down one HTTP server side runs
more that twice as slowly as 120 users balanced over the two machines.
Anyone have any ( polite :-) suggestions or comments, or is it just one of
those wierd things that happens from time to time?

Peter.

[Sven Vermeulen]

Post by Pete Bennett
I just cannot imagine why the 60 workload down one HTTP server side runs
more that twice as slowly as 120 users balanced over the two machines.
Anyone have any ( polite :-) suggestions or comments, or is it just one of
those wierd things that happens from time to time?

Where do you notice the slow-down? Are the websphere response times
already large or is it TAMeB that's slow?

Perhaps the SSL Session ID cache in the first situation is too limited
while in the second situation the users rotate over a large enough
timescale not to see any serious effects on the SSL Session ID cache?

Wkr,
Sven Vermeulen