IAA Compression Batching:
=========================
This patch-series introduces the use of the Intel Analytics Accelerator
(IAA) for parallel compression of pages in large folios.
The patch-series is organized as follows:
1) crypto acomp & iaa_crypto driver enablers for batching: Relevant
patches are tagged with "crypto:" in the subject:
Patch 1) New acomp_alg/crypto_acomp batch_compress() and batch_decompress()
interfaces, that swap modules can invoke using the new
batching API crypto_acomp_batch_compress() and
crypto_acomp_batch_decompress().
Patch 2) New CRYPTO_ACOMP_REQ_POLL acomp_req flag to act as a gate for
async poll mode in iaa_crypto.
Patch 3) iaa-crypto driver implementations for async polling,
crypto_acomp_batch_compress() and crypto_acomp_batch_decompress().
Patch 4) Modifying the default iaa_crypto driver mode to async.
Patch 5) Disabling verify_compress by default, to facilitate users to
run IAA easily for comparison with software compressors.
Patch 6) Changing the cpu-to-iaa mappings to more evenly balance cores
to IAA devices.
Patch 7) Addition of a "global_wq" per IAA, which can be used as a
global resource for compress jobs for the socket. If the user
configures 2WQs per IAA device, the driver will distribute
compress jobs from all cores on the socket to the "global_wqs"
of all the IAA devices on that socket, in a round-robin
manner. This can be used to improve compression throughput for
workloads that see a lot of swapout activity.
2) zswap modifications to enable compress batching in zswap_store() of
large folios (including pmd-mappable folios):
Patch 8) acomp_ctx mutex lock acquire/release once optimizations in
zswap_store() and a minor change in releasing the lock in
zswap_decompress().
Patch 9) Change the "struct crypto_acomp_ctx" to contain a configurable
number of acomp_reqs and buffers.
Patch 10) Introduce a separate per-cpu "acomp_batch_ctx" member in
"struct zswap_pool" to be able to allocate multiple
acomp_reqs/buffers for use in batching, as needed, per core.
Patch 11) Allocation of the per-cpu "acomp_batch_ctx" for a
zswap_pool.
Patch 12) Add a new "sysctl vm.compress-batching" 0/1 switch to
enable/disable compress batching dynamically at runtime.
Patch 13) zswap_store() IAA compress batching implementation with
minimal memory footprint cost per-cpu, and using the new
crypto_acomp_batch_compress() iaa_crypto driver API.
With the runtime configuration to enable compress batching and the crypto
batching API added in v2, this feature will be enabled only on Intel
platforms that have IAA.
System setup for testing:
=========================
Testing of this patch-series was done with mm-unstable as of 10-29-2024,
commit 9fb8e0a1c486, without and with this patch-series.
Data was gathered on an Intel Sapphire Rapids server, dual-socket 56 cores
per socket, 4 IAA devices per socket, 503 GiB RAM and 525G SSD disk
partition swap. Core frequency was fixed at 2500MHz.
Other kernel configuration parameters:
zswap compressor : zstd, deflate-iaa
zswap allocator : zsmalloc
vm.page-cluster : 0, 2
IAA "compression verification" is disabled and IAA is run in the the async
poll mode (the defaults with this series). 2WQs are configured per IAA
device. Compress jobs from all cores on a socket are distributed among all
4 IAA devices on the same socket.
I ran experiments with these workloads:
1) usemem 30 processes with these large folios enabled to "always":
- 16k/32k/64k
- 2048k
2) Kernel compilation allmodconfig with 2G max memory, run in tmpfs with
these large folios enabled to "always":
- 16k/32k/64k
Performance testing (usemem30):
===============================
The vm-scalability "usemem" test was run in a cgroup whose memory.high
was fixed at 150G. The is no swap limit set for the cgroup. 30 usemem
processes were run, each allocating and writing 10G of memory, and sleeping
for 10 sec before exiting:
usemem --init-time -w -O -s 10 -n 30 10g
16k/32/64k folios: usemem30/deflate-iaa:
========================================
-------------------------------------------------------------------------------
mm-unstable-10-29-2024 v2 of this patch-series
-------------------------------------------------------------------------------
zswap compressor deflate-iaa deflate-iaa deflate-iaa
vm.compress-batching n/a 0 1
vm.page-cluster 2 2 2
-------------------------------------------------------------------------------
Total throughput (KB/s) 7,756,632 7,753,984 8,075,817
Avg throughput (KB/s) 258,554 258,466 269,193
elapsed time (sec) 87.75 88.71 85.82
sys time (sec) 2,073.04 2,147.47 2,030.52
-------------------------------------------------------------------------------
memcg_high 715,854 714,238 720,459
memcg_swap_fail 1,194 1,175 1,250
zswpout 64,510,869 64,510,832 64,511,219
zswpin 458 456 450
pswpout 0 0 0
pswpin 0 0 0
thp_swpout 0 0 0
thp_swpout_fallback 0 0 0
16kB-mthp_swpout_fallback 0 0 0
32kB-mthp_swpout_fallback 0 0 0
64kB-mthp_swpout_fallback 1,194 1,175 1,250
pgmajfault 3,183 3,513 3,116
swap_ra 108 125 116
swap_ra_hit 45 65 43
ZSWPOUT-16kB 2 3 3
ZSWPOUT-32kB 1 1 2
ZSWPOUT-64kB 4,030,658 4,030,672 4,030,624
SWPOUT-16kB 0 0 0
SWPOUT-32kB 0 0 0
SWPOUT-64kB 0 0 0
-------------------------------------------------------------------------------
16k/32/64k folios: usemem30/zstd:
=================================
-------------------------------------------------------------------------------
mm-unstable-10-29-2024 v2 of this patch-series
-------------------------------------------------------------------------------
zswap compressor zstd zstd
vm.compress-batching n/a 0
vm.page-cluster 2 2
-----------------------------------------------------------------------------
Total throughput (KB/s) 6,054,147 6,109,360
Avg throughput (KB/s) 201,804 203,645
elapsed time (sec) 111.66 111.72
sys time (sec) 2,693.21 2,685.27
-----------------------------------------------------------------------------
memcg_high 489,133 480,524
memcg_swap_fail 1,045 1,308
zswpout 48,931,716 48,931,540
zswpin 407 394
pswpout 0 0
pswpin 0 0
thp_swpout 0 0
thp_swpout_fallback 0 0
16kB-mthp_swpout_fallback 0 0
32kB-mthp_swpout_fallback 0 0
64kB-mthp_swpout_fallback 1,045 1,308
pgmajfault 3,095 3,424
swap_ra 136 101
swap_ra_hit 86 50
ZSWPOUT-16kB 2 4
ZSWPOUT-32kB 0 2
ZSWPOUT-64kB 3,057,161 3,056,927
SWPOUT-16kB 0 0
SWPOUT-32kB 0 0
SWPOUT-64kB 0 0
-----------------------------------------------------------------------------
2M folios: usemem30/deflate-iaa:
================================
-------------------------------------------------------------------------------
mm-unstable-10-29-2024 v2 of this patch-series
-------------------------------------------------------------------------------
zswap compressor deflate-iaa deflate-iaa deflate-iaa
vm.compress-batching n/a 0 1
vm.page-cluster 2 2 2
-------------------------------------------------------------------------------
Total throughput (KB/s) 7,948,345 8,096,440 8,165,171
Avg throughput (KB/s) 264,944 269,881 272,172
elapsed time (sec) 88.18 87.13 87.30
sys time (sec) 2,067.56 2,018.08 2,046.79
-------------------------------------------------------------------------------
memcg_high 91,002 87,243 92,084
memcg_swap_fail 39 56 54
zswpout 64,518,833 64,520,439 64,520,116
zswpin 413 452 504
pswpout 0 0 0
pswpin 0 0 0
thp_swpout 0 0 0
thp_swpout_fallback 39 56 54
2048kB-mthp_swpout_fallback 39 56 54
pgmajfault 10,946 15,737 9,645
swap_ra 23,456 36,495 19,247
swap_ra_hit 23,406 36,431 19,193
ZSWPOUT-2048kB 125,915 125,913 125,912
SWPOUT-2048kB 0 0 0
-------------------------------------------------------------------------------
2M folios: usemem30/zstd:
=========================
-------------------------------------------------------------------------------
mm-unstable-10-29-2024 v2 of this patch-series
-------------------------------------------------------------------------------
zswap compressor zstd zstd
vm.compress-batching n/a 0
vm.page-cluster 2 2
-----------------------------------------------------------------------------
Total throughput (KB/s) 6,300,116 6,278,179
Avg throughput (KB/s) 210,003 209,272
elapsed time (sec) 110.21 111.72
sys time (sec) 2,504.45 2,542.59
-----------------------------------------------------------------------------
memcg_high 57,036 60,090
memcg_swap_fail 61 50
zswpout 48,934,256 48,904,582
zswpin 387 380
pswpout 0 0
pswpin 0 0
thp_swpout 0 0
thp_swpout_fallback 61 50
2048kB-mthp_swpout_fallback 61 50
pgmajfault 3,713 6,146
swap_ra 2,004 8,133
swap_ra_hit 1,960 8,088
ZSWPOUT-2048kB 95,511 95,460
SWPOUT-2048kB 0 0
-----------------------------------------------------------------------------
Performance testing (Kernel compilation, allmodconfig):
=======================================================
The experiments with kernel compilation test in tmpfs use the
"allmodconfig" that takes ~12 minutes, and has considerable swapout
activity. The cgroup's memory.max is set to 2G.
16k/32k/64k folios: Kernel compilation/allmodconfig/deflate-iaa:
================================================================
-------------------------------------------------------------------------------
mm-unstable-10-29-2024 v2 of this patch-series
-------------------------------------------------------------------------------
zswap compressor deflate-iaa deflate-iaa deflate-iaa
vm.compress-batching n/a 0 1
vm.page-cluster 0 0 0
-------------------------------------------------------------------------------
real_sec 801.25 790.87 768.92
user_sec 15,776.31 15,755.97 15,753.89
sys_sec 4,250.34 3,877.02 3,892.17
Max_Res_Set_Size_KB 1,869,428 1,873,376 1,871,600
-------------------------------------------------------------------------------
memcg_high 0 0 0
memcg_swap_fail 0 0 0
zswpout 106,798,327 105,469,307 104,528,841
zswpin 31,542,093 30,469,671 30,596,840
pswpout 774 290 80
pswpin 370 288 59
thp_swpout 0 0 0
thp_swpout_fallback 0 0 0
16kB-mthp_swpout_fallback 0 0 0
32kB-mthp_swpout_fallback 0 0 0
64kB-mthp_swpout_fallback 16,340 12,633 12,000
pgmajfault 33,983,602 32,783,214 32,731,862
swap_ra 0 0 0
swap_ra_hit 1,467 5,112 3,854
ZSWPOUT-16kB 1,475,121 1,435,571 1,426,738
ZSWPOUT-32kB 821,119 813,202 790,658
ZSWPOUT-64kB 3,483,295 3,490,244 3,435,056
SWPOUT-16kB 1 0 0
SWPOUT-32kB 3 0 0
SWPOUT-64kB 40 18 4
-------------------------------------------------------------------------------
16k/32k/64k folios: Kernel compilation/allmodconfig/zstd:
=========================================================
-------------------------------------------------------------------------------
mm-unstable-10-29-2024 v2 of this patch-series
-------------------------------------------------------------------------------
zswap compressor zstd zstd
vm.compress-batching n/a 0
vm.page-cluster 0 0
-------------------------------------------------------------------------------
real_sec 812.38 800.09
user_sec 15,774.12 15,771.02
sys_sec 5,283.64 5,257.05
Max_Res_Set_Size_KB 1,872,688 1,873,444
-------------------------------------------------------------------------------
memcg_high 0 0
memcg_swap_fail 0 0
zswpout 91,540,018 90,338,507
zswpin 26,421,271 26,485,837
pswpout 64 144
pswpin 64 114
thp_swpout 0 0
thp_swpout_fallback 0 0
16kB-mthp_swpout_fallback 0 0
32kB-mthp_swpout_fallback 0 0
64kB-mthp_swpout_fallback 4,509 566
pgmajfault 28,341,722 28,427,509
swap_ra 0 0
swap_ra_hit 3,359 2,931
ZSWPOUT-16kB 1,287,206 1,266,947
ZSWPOUT-32kB 707,746 700,270
ZSWPOUT-64kB 2,985,002 2,940,288
SWPOUT-16kB 0 0
SWPOUT-32kB 0 0
SWPOUT-64kB 4 9
-------------------------------------------------------------------------------
Summary:
========
The performance testing data with usemem 30 processes and kernel
compilation test show throughput gains and elapsed/sys time reduction with
zswap_store() large folios using IAA compress batching.
The iaa_crypto wq stats will show almost the same number of compress calls
for wq.1 of all IAA devices. wq.0 will handle decompress calls exclusively.
We see a latency reduction of 2.5% by distributing compress jobs among all
IAA devices on the socket (based on v1 data).
We can expect to see even more significant performance and throughput
improvements if we use the parallelism offered by IAA to batch compress the
pages comprising a batch of 4K (really any-order) folios, not just batching
within large folios. This is the reclaim batching patch 13 in v1,
which will be submitted in a separate patch-series.
Our internal validation of IAA compress/decompress batching in highly
contended Sapphire Rapids server setups with workloads running on 72 cores
for ~25 minutes under stringent memory limit constraints have shown up to
50% reduction in sys time and 3.5% reduction in workload run time as
compared to software compressors.
Changes since v2:
=================
1) Rebased to mm-unstable as of 11-5-2024, commit 7994b7ea6ac8.
2) Fixed an issue in zswap_create_acomp_ctx() with checking for NULL
returned by kmalloc_node() for acomp_ctx->buffers and for
acomp_ctx->reqs.
3) Fixed a bug in zswap_pool_can_batch() for returning true if
pool->can_batch_comp is found to be equal to BATCH_COMP_ENABLED, and if
the per-cpu acomp_batch_ctx tests true for batching resources having
been allocated on this cpu. Also, changed from per_cpu_ptr() to
raw_cpu_ptr().
4) Incorporated the zswap_store_propagate_errors() compilation warning fix
suggested by Dan Carpenter. Thanks Dan!
5) Replaced the references to SWAP_CRYPTO_SUB_BATCH_SIZE in comments in
zswap.h, with SWAP_CRYPTO_BATCH_SIZE.
Changes since v1:
=================
1) Rebased to mm-unstable as of 11-1-2024, commit 5c4cf96cd702.
2) Incorporated Herbert's suggestions to use an acomp_req flag to indicate
async/poll mode, and to encapsulate the polling functionality in the
iaa_crypto driver. Thanks Herbert!
3) Incorporated Herbert's and Yosry's suggestions to implement the batching
API in iaa_crypto and to make its use seamless from zswap's
perspective. Thanks Herbert and Yosry!
4) Incorporated Yosry's suggestion to make it more convenient for the user
to enable compress batching, while minimizing the memory footprint
cost. Thanks Yosry!
5) Incorporated Yosry's suggestion to de-couple the shrink_folio_list()
reclaim batching patch from this series, since it requires a broader
discussion.
Requesting the maintainers & reviewers to kindly review v3 of this
patch-series instead of v2.
I would greatly appreciate code review comments for the iaa_crypto driver
and mm patches included in this series!
Thanks,
Kanchana
Kanchana P Sridhar (13):
crypto: acomp - Define two new interfaces for compress/decompress
batching.
crypto: iaa - Add an acomp_req flag CRYPTO_ACOMP_REQ_POLL to enable
async mode.
crypto: iaa - Implement compress/decompress batching API in
iaa_crypto.
crypto: iaa - Make async mode the default.
crypto: iaa - Disable iaa_verify_compress by default.
crypto: iaa - Change cpu-to-iaa mappings to evenly balance cores to
IAAs.
crypto: iaa - Distribute compress jobs to all IAA devices on a NUMA
node.
mm: zswap: acomp_ctx mutex lock/unlock optimizations.
mm: zswap: Modify struct crypto_acomp_ctx to be configurable in nr of
acomp_reqs.
mm: zswap: Add a per-cpu "acomp_batch_ctx" to struct zswap_pool.
mm: zswap: Allocate acomp_batch_ctx resources for a given zswap_pool.
mm: Add sysctl vm.compress-batching switch for compress batching
during swapout.
mm: zswap: Compress batching with Intel IAA in zswap_store() of large
folios.
crypto/acompress.c | 2 +
drivers/crypto/intel/iaa/iaa_crypto_main.c | 717 +++++++++++++++--
include/crypto/acompress.h | 87 +++
include/crypto/internal/acompress.h | 16 +
include/linux/mm.h | 2 +
include/linux/zswap.h | 91 +++
kernel/sysctl.c | 9 +
mm/swap.c | 6 +
mm/zswap.c | 865 +++++++++++++++++++--
9 files changed, 1701 insertions(+), 94 deletions(-)
base-commit: 7994b7ea6ac880efd0c38fedfbffd5ab8b1b7b2b
--
2.27.0
