On Wednesday, 07/23/2008 at 04:15 EDT, "McKown, John" <[EMAIL PROTECTED]> wrote:
> Just some thoughts on this. From what I understand, a hipersocket is > really a way to "move" data from one memory location in one LPAR to > another memory location on another LPAR in the same CEC. Or to the same LPAR, it doesn't matter. > Now, is that > "move" done by a CP? Or is it done by the SAP? If it is done by a CP, > then it is "knee-capped" if the CP is "knee-capped". If it is done by a > SAP, then it is not. I would hope that it is actually done by a SAP. But > that would mean that the speed could be influenced by how busy the SAP > is doing other things. Given that CP's are millicoded, I think you cannot presume that internal speed and apparent speed are the same. HiperSockets is internal to the CPU, using iQDIO interfaces, and is synchronous with respect to data movement. That is, the data is not stored "somewhere else". So, only if they actually turn down the clock speed would it necessarily be slower. Hence, measure it and find out if *your* workload on *your* system would benefit. Alan Altmark z/VM Development IBM Endicott
