On 06/11/2012 09:06 AM, Mel Gorman wrote:
My initial support for this patch was based on an artifical load but one I felt was plausible to trigger if CMA was being used. In a normal workload I thought it might be possible to hit if a large process exited freeing a lot of pagetable pages from MIGRATE_UNMOVABLE blocks at the same time but that is a little unlikely and a test case would also look very artifical. Hence, I believe that if you require a real workload to demonstrate the benefit of the patch that it will be very difficult to find. The primary decision is if CMA needs this or not. I was under the impression that it was a help for CMA allocation success rates but I may be mistaken.
If it helps CMA allocation rates, it should also help allocation rates for transparent hugepages. Conveniently, THP allocation rates are already exported in /proc/vmstat. Now all we need is a test load :) -- All rights reversed -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>