You need the full picture from you client – if he is the one responsible for monitoring, he should be able to supply every machine’s CPU, Memory, Network and Disk stats. I would also ask for CPU measure ments for every CPU on the machine, windows %CPU is an average, with multiple CPUs I would want to know each one was behaving as expected.

Given the above plus application specific stats (like the IIS ones you discussed) – you should be able to see what’s inbalanced, overloaded or not busy at all.

We package up system monitoring with performance testing, there are various benefits and reasons for this – but when when performance is bad one of the usual next questions is “what did the machines look like at the time”.

There should be a monitoring stat which validates you have applied the planned load. This could be a network interface to a load balancer.

It is possible that the load test data or nature of the test tool (range of load injector IP addresses) are causing the system to be imbalanced. I’m trying to think of a good example, let’s say user ids are divided alphabetically and each web server deals with its allocated range of user ids… anything is possible!

Regards,

Stephen.