On Thu, Nov 09, 2023 at 01:20:29PM +0800, Huang, Ying wrote:
> Byungchul Park <byungchul@xxxxxx> writes:
>
> > Hi everyone,
> >
> > While I'm working with CXL memory, I have been facing migration overhead
> > esp. TLB shootdown on promotion or demotion between different tiers.
> > Yeah.. most TLB shootdowns on migration through hinting fault can be
> > avoided thanks to Huang Ying's work, commit 4d4b6d66db ("mm,unmap: avoid
> > flushing TLB in batch if PTE is inaccessible").
> >
> > However, it's only for ones using hinting fault. I thought it'd be much
> > better if we have a general mechanism to reduce # of TLB flushes and
> > TLB misses, that we can apply to any type of migration. I tried it only
> > for tiering migration for now tho.
> >
> > I'm suggesting a mechanism to reduce TLB flushes by keeping source and
> > destination of folios participated in the migrations until all TLB
> > flushes required are done, only if those folios are not mapped with
> > write permission PTE entries at all. I worked Based on v6.6-rc5.
> >
> > Can you believe it? I saw the number of TLB full flush reduced about
> > 80% and iTLB miss reduced about 50%, and the time wise performance
> > always shows at least 1% stable improvement with the workload I tested
> > with, XSBench. However, I believe that it would help more with other
> > ones or any real ones. It'd be appreciated to let me know if I'm missing
> > something.
>
> Can you help to test the effect of commit 7e12beb8ca2a ("migrate_pages:
> batch flushing TLB") for your test case? To test it, you can revert it
> and compare the performance before and after the reverting.
>
> And, how do you trigger migration when testing XSBench? Use a tiered
> memory system, and migrate pages between DRAM and CXL memory back and
> forth? If so, how many pages will you migrate for each migration
It was not an actual CXL memory but a cpuless remote numa node's DRAM
recognized as a slow tier (node_is_toptier() == false) by the kernel.
It's been okay to me because I've been focusing on TLB # and migration #
while working with numa tiering mechanism and, I think, the time wise
performance will be followed, big or little depending on the system
configuration.
So it migrates pages between the two DRAMs back and forth - promotion by
hinting fault and demotion by page reclaim. I tested what you asked me
with another slower system to make TLB miss overhead stand out.
Unfortunately I got even worse result with vanilla v6.6-rc5 than
v6.6-rc5 with 7e12beb8ca2a reverted, while the 'v6.6-rc5 + migrc'
definitely shows far better result.
Thoughts?
Byungchul
---
Architecture - x86_64
QEMU - kvm enabled, host cpu
Numa - 2 nodes (16 CPUs 1GB, no CPUs 8GB)
Kernel - v6.6-rc5, NUMA_BALANCING_MEMORY_TIERING, demotion enabled
Benchmark - XSBench -p 50000000 (-p option makes the runtime longer)
CASE1 - mainline v6.6-rc5 + 7e12beb8ca2a reverted
-------------------------------------------------
$ perf stat -a \
-e itlb.itlb_flush \
-e tlb_flush.dtlb_thread \
-e tlb_flush.stlb_any \
-e dTLB-load-misses \
-e dTLB-store-misses \
-e iTLB-load-misses \
./XSBench -p 50000000
Performance counter stats for 'system wide':
190247118 itlb.itlb_flush
716182438 tlb_flush.dtlb_thread
327051673 tlb_flush.stlb_any
119542331968 dTLB-load-misses
724072795 dTLB-store-misses
3054343419 iTLB-load-misses
1172.580552728 seconds time elapsed
$ cat /proc/vmstat
...
numa_pages_migrated 5968431
pgmigrate_success 12484773
nr_tlb_remote_flush 6614459
nr_tlb_remote_flush_received 96022799
nr_tlb_local_flush_all 50869
nr_tlb_local_flush_one 785597
...
CASE2 - mainline v6.6-rc5 (vanilla)
-------------------------------------------------
$ perf stat -a \
-e itlb.itlb_flush \
-e tlb_flush.dtlb_thread \
-e tlb_flush.stlb_any \
-e dTLB-load-misses \
-e dTLB-store-misses \
-e iTLB-load-misses \
./XSBench -p 50000000
Performance counter stats for 'system wide':
55139061 itlb.itlb_flush
286725687 tlb_flush.dtlb_thread
199687660 tlb_flush.stlb_any
119497951269 dTLB-load-misses
358434759 dTLB-store-misses
1867135967 iTLB-load-misses
1181.311084373 seconds time elapsed
$ cat /proc/vmstat
...
numa_pages_migrated 8190027
pgmigrate_success 17098994
nr_tlb_remote_flush 1955114
nr_tlb_remote_flush_received 29028093
nr_tlb_local_flush_all 140921
nr_tlb_local_flush_one 740767
...
CASE3 - mainline v6.6-rc5 + migrc
-------------------------------------------------
$ perf stat -a \
-e itlb.itlb_flush \
-e tlb_flush.dtlb_thread \
-e tlb_flush.stlb_any \
-e dTLB-load-misses \
-e dTLB-store-misses \
-e iTLB-load-misses \
./XSBench -p 50000000
Performance counter stats for 'system wide':
6337091 itlb.itlb_flush
157229778 tlb_flush.dtlb_thread
148240163 tlb_flush.stlb_any
117701381319 dTLB-load-misses
231212468 dTLB-store-misses
973083466 iTLB-load-misses
1105.756705157 seconds time elapsed
$ cat /proc/vmstat
...
numa_pages_migrated 8791934
pgmigrate_success 18276174
nr_tlb_remote_flush 311146
nr_tlb_remote_flush_received 4387708
nr_tlb_local_flush_all 143883
nr_tlb_local_flush_one 740953
...
