This series reduces scan rates and success rates of compaction, primarily by using the free lists to shorten scans, better controlling of skip information and whether multiple scanners can target the same block and capturing pageblocks before being stolen by parallel requests. The series is based on the 4.21/5.0 merge window after Andrew's tree had been merged. It's known to rebase cleanly. Primarily I'm using thpscale to measure the impact of the series. The benchmark creates a large file, maps it, faults it, punches holes in the mapping so that the virtual address space is fragmented and then tries to allocate THP. It re-executes for different numbers of threads. From a fragmentation perspective, the workload is relatively benign but it does stress compaction. The overall impact on latencies for a 1-socket machine is baseline patches Amean fault-both-3 5362.80 ( 0.00%) 4446.89 * 17.08%* Amean fault-both-5 9488.75 ( 0.00%) 5660.86 * 40.34%* Amean fault-both-7 11909.86 ( 0.00%) 8549.63 * 28.21%* Amean fault-both-12 16185.09 ( 0.00%) 11508.36 * 28.90%* Amean fault-both-18 12057.72 ( 0.00%) 19013.48 * -57.69%* Amean fault-both-24 23939.95 ( 0.00%) 19676.16 * 17.81%* Amean fault-both-30 26606.14 ( 0.00%) 27363.23 ( -2.85%) Amean fault-both-32 31677.12 ( 0.00%) 23154.09 * 26.91%* While there is a glitch at the 18-thread mark, it's known that the base page allocation latency was much lower and huge pages were taking longer -- partially due a high allocation success rate. The allocation success rates are much improved baseline patches Percentage huge-3 70.93 ( 0.00%) 98.30 ( 38.60%) Percentage huge-5 56.02 ( 0.00%) 83.36 ( 48.81%) Percentage huge-7 60.98 ( 0.00%) 89.04 ( 46.01%) Percentage huge-12 73.02 ( 0.00%) 94.36 ( 29.23%) Percentage huge-18 94.37 ( 0.00%) 95.87 ( 1.58%) Percentage huge-24 84.95 ( 0.00%) 97.41 ( 14.67%) Percentage huge-30 83.63 ( 0.00%) 96.69 ( 15.62%) Percentage huge-32 81.69 ( 0.00%) 96.10 ( 17.65%) That's a nearly perfect allocation success rate. The biggest impact is on the scan rates Compaction migrate scanned 106520811 26934599 Compaction free scanned 4180735040 26584944 The number of pages scanned for migration was reduced by 74% and the free scanner was reduced by 99.36%. So much less work in exchange for lower latency and better success rates. The series was also evaluated using a workload that heavily fragments memory but the benefits there are also significant, albeit not presented. It was commented that we should be rethinking scanning entirely and to a large extent I agree. However, to achieve that you need a lot of this series in place first so it's best to make the linear scanners as best as possible before ripping them out. include/linux/compaction.h | 3 +- include/linux/gfp.h | 7 +- include/linux/mmzone.h | 2 + include/linux/sched.h | 4 + kernel/sched/core.c | 3 + mm/compaction.c | 1031 ++++++++++++++++++++++++++++++++++---------- mm/internal.h | 23 +- mm/migrate.c | 2 +- mm/page_alloc.c | 70 ++- 9 files changed, 908 insertions(+), 237 deletions(-) -- 2.16.4
