Trying to find some info on how the cores communicate with a Dual Processor, Quad Xeon E5420. We have a code that uses MPI to measure the speed that each core communicates with another.
Between machines we have one consistent speed, call it "internet speed" say. However inside a machine, we get 4 "internet speed" communications, 2 slow communications and one fast communication. (Obviously we do not check how fast a core communicated with itself). The differences are on the order of 20-30% and repeatable over machines. Anyone have any idea why or how this is happening? Kind of useful for load balancing issues.
