Les Mikesell wrote: > Wouldn't it also make sense to compare the threaded version against > running n instances of the unthreaded version where n= the number of > CPUs available? >
It might be interesting, sure, but it's not quite apples-to-apples since those are n separate caches that don't interact with one another at all, whereas an n-threaded instance is a single interface to a single cache. Also, that requires a different code path on the client side (multigets will need to be split among multiple servers) which might affect the benchmark results. But sure, wouldn't hurt to see that as well. -Steve
