On 03/29/2018 06:47 AM, Peter Zijlstra wrote: > Further I think Dave argued that we should not change the llc-size, > because while SNC presents a subset of the cache to local CPUs, for > remote data the whole cache is still available, again something some > applications might rely on.
BTW, I may have argued this in the past, but I don't think it's the best thing to do. If anything, we should optimize for the _users_ of this information: the performance-sensitive ones who are digging up the cache topology. They are also likely to be the most NUMA-affinitized and stay node-local with much of their memory traffic. That would seem to point us in the direction of enumerating two separate, half-sized LLCs "shared" only by the slice when SNC mode is on. That's what I've argued to the hardware folks lately, at least.
