Messages in this thread | | | Date | Mon, 16 May 2005 14:21:01 -0700 | From | "Martin J. Bligh" <> | Subject | Re: NUMA aware slab allocator V3 |
| |
> Right, the SGI boxes have discontiguous memory within a node, but it's > not represented by pgdats (like you said, one 'virtual memmap' spans > the whole address space of a node). Sparse can help simplify this > across platforms, but has the potential to be more expensive for > systems with dynamically sized holes, due to the additional calculation > and potential cache miss associated with indexing into the correct > memmap (Dave can probably correct me here, it's been awhile). With a > virtual memmap, you only occasionally take a TLB miss on the struct > page access after indexing into the array.
That's exactly what was brilliant about Andy's code ... it fixed that, there shouldn't be extra references ...
>> transition config options are a bit of a mess ... Andy, I presume >> CONFIG_NEED_MULTIPLE_NODES is really CONFIG_NEED_MULTIPLE_PGDATS ? > > Yeah, makes sense for the NUMA aware slab allocator to depend on > CONFIG_NUMA.
Andy confirmed offline that this is really CONFIG_NEED_MULTIPLE_PGDATS, and is just named wrong.
M.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |