I have this question. In a 10.2.1 server site with a default cluster and 2 machines with 2 Web Adaptors (being load balanced) we are seeing one GIS server being strapped at 100% CPU while the second runs at about 10% CPU. My question is how the server architecture handles spreading the load. It was my understanding that the site is an architecture of peers and that the process of request handling was this:
The web adaptors send requests to each machine in the site / cluster in the order that it knows about them. If gis server 1 (or the one receiving the request) is busy the request goes back to web adapter then down to gis server 2. If both are busy then the request cues at gis server1. What we are seeing though is that during load testing gis server 1 is maxing out at 100% cpu and the second server is sitting at about 15%. This is for requests to the same map service. So the expectation would be that each machine would have similar load? It seems that the WAs are sending more requests to the first node than the other.
Am I missing a piece of the algorithm puzzle?
Also is there any configuration available on the WAs to say how much load a server must have before it will accept a request - also can we do this at the server peer level?