Messages in this thread | | | Date | Fri, 29 Oct 2010 14:33:02 -0500 (CDT) | From | Christoph Lameter <> | Subject | Re: [PATCH 1/2] mm: page allocator: Adjust the per-cpu counter threshold when memory is low |
| |
On Fri, 29 Oct 2010, Andrew Morton wrote:
> > Doing that caused cache misses in the past and reduced the performance of > > the ZVCs. This way the threshold is in the same cacheline as the > > differentials. > > This sounds wrong. As long as that threshold isn't stored in a > cacheline which other CPUs are modifying, all CPUs should be able to > happily cache it. Maybe it needed a bit of padding inside the zone > struct.
High speed cpu caches are a very scarce resource. The differentials are not in the zone struct. Tried to put it onto a single cacheline. Even that did not do the trick for the large configurations.
The same optimizations are done in the slab allocators by the way.
Use of the percpu_counter() would at least quadruple the cache footprint in use for the counters. percpu_counters() is using s32 and not s8.
| |