[PATCH v2 0/8] mm/swap: Regular page swap optimizations
From: Tim Chen
Date: Thu Oct 20 2016 - 19:32:34 EST
We appreciate feedback about this patch series from the
community. Historically, neither the performance nor latency of the swap
path mattered. The underlying I/O was slow enough to hide any latency
coming from software and the low IOPS kept the overall CPU impact low.
Times have changed. Coming generation of Solid state Block device
latencies are getting down to sub 100 usec, which is within an order of
magnitude of DRAM, and their performance is orders of magnitude higher
than the single- spindle rotational media we've swapped to historically.
This could benefit many usage scenearios. For example cloud providers who
overcommit their memory (as VM don't use all the memory provisioned).
Having a fast swap will allow them to be more aggressive in memory
overcommit and fit more VMs to a platform.
In our testing [see footnote], the median latency that the
kernel adds to a page fault is 15 usec, which comes quite close
to the amount that will be contributed by the underlying I/O
The software latency comes mostly from contentions on the locks
protecting the radix tree of the swap cache and also the locks protecting
the individual swap devices. The lock contentions already consumed
35% of cpu cycles in our test. In the very near future,
software latency will become the bottleneck to swap performnace as
block device I/O latency gets within the shouting distance of DRAM speed.
This patch set, plus a previous patch Ying already posted
(commit: f6498b3f) reduced the median page fault latency
from 15 usec to 4 usec (375% reduction) for DRAM based pmem
Patch 1 is a clean up patch.
Patch 2 creates a lock per cluster, this gives us a more fine graind lock
that can be used for accessing swap_map, and not lock the whole
Patch 3 splits the swap cache radix tree into 64MB chunks, reducing
the rate that we have to contende for the radix tree.
Patch 4 eliminates unnecessary page allocation for read ahead.
Patch 5-8 create a per cpu cache of the swap slots, so we don't have
to contend on the swap device to get a swap slot or to release
a swap slot. And we allocate and release the swap slots
in batches for better efficiency.
Ying Huang & Tim Chen
Testing was done on 4.8-rc3-mm1 kernel with/without optimizations from
this patche series plus one additional patch Ying posted earlier on
removing radix tree write back tag in swap cache. Eight threads performed
random memory access on a 2 socket Haswell using swap mounted on RAM
based PMEM block device. This emulated a moderate load and a SWAP
device unbounded by I/O speed. The aggregate working set is twice the
RAM size. We instrumented the kernel to measure the page fault latency.
1. Fix bug in the index limit used in scan_swap_map_try_ssd_cluster
when searching for empty slots in cluster.
2. Fix bug in swap off that incorrectly determines if we still have
swap devices left.
3. Port patches to mmotm-2016-10-11-15-46 branch
Huang, Ying (3):
mm/swap: Fix kernel message in swap_info_get()
mm/swap: Add cluster lock
mm/swap: Split swap cache into 64MB trunks
Tim Chen (5):
mm/swap: skip read ahead for unreferenced swap slots
mm/swap: Allocate swap slots in batches
mm/swap: Free swap slots in batch
mm/swap: Add cache for swap slots allocation
mm/swap: Enable swap slots cache usage
include/linux/swap.h | 35 ++-
include/linux/swap_slots.h | 37 +++
mm/Makefile | 2 +-
mm/swap.c | 6 -
mm/swap_slots.c | 306 +++++++++++++++++++++++++
mm/swap_state.c | 76 +++++-
mm/swapfile.c | 558 +++++++++++++++++++++++++++++++++++----------
7 files changed, 875 insertions(+), 145 deletions(-)
create mode 100644 include/linux/swap_slots.h
create mode 100644 mm/swap_slots.c