On Sun, Jul 31, 2016 at 11:47 AM, EduRam <eduardo.rama...@gmail.com> wrote:
> It appears it will be possible to allocate partial CPU L3 cache to > processes. If I remember correctly, the original idea for the CAT technology was to have "cache smashing" workloads on the same machines as other workloads. The problem is that one workload fills up the cache and then when another workload comes in, it has to pull everything into cache again. It was a major reason to split work over multiple machines, and it really hurts in virtualization environments. My intuition somewhat follows what rlh@ writes: isolating a marking phase over a heap so it can't evict every cache line sounds quite intriguing. In fact, any "sweep" (not in the GC sense) over data could end up evicting lots of useful cache lines which are accessed right after. -- J. -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
