Skip to content

L
linux
Commit 2fcbc413 authored by Mike Rapoport's avatar Mike Rapoport Committed by Jonathan Corbet
Browse files
Options

docs/vm: ksm.txt: convert to ReST format 

Signed-off-by: default avatarMike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: default avatarJonathan Corbet <corbet@lwn.net>
parent e3f2025a
Hide whitespace changes
Inline Side-by-side
Showing with 110 additions and 105 deletions
Documentation/vm/ksm.txt
View file @ 2fcbc413
How to use the Kernel Samepage Merging feature .. _ksm:
----------------------------------------------
=======================
Kernel Samepage Merging
=======================
KSM is a memory-saving de-duplication feature, enabled by CONFIG_KSM=y, KSM is a memory-saving de-duplication feature, enabled by CONFIG_KSM=y,
added to the Linux kernel in 2.6.32. See mm/ksm.c for its implementation, added to the Linux kernel in 2.6.32. See ``mm/ksm.c`` for its implementation,
and http://lwn.net/Articles/306704/ and http://lwn.net/Articles/330589/ and http://lwn.net/Articles/306704/ and http://lwn.net/Articles/330589/
The KSM daemon ksmd periodically scans those areas of user memory which The KSM daemon ksmd periodically scans those areas of user memory which
...@@ -51,110 +54,112 @@ Applications should be considerate in their use of MADV_MERGEABLE, ...@@ -51,110 +54,112 @@ Applications should be considerate in their use of MADV_MERGEABLE,
restricting its use to areas likely to benefit. KSM's scans may use a lot restricting its use to areas likely to benefit. KSM's scans may use a lot
of processing power: some installations will disable KSM for that reason. of processing power: some installations will disable KSM for that reason.
The KSM daemon is controlled by sysfs files in /sys/kernel/mm/ksm/, The KSM daemon is controlled by sysfs files in ``/sys/kernel/mm/ksm/``,
readable by all but writable only by root: readable by all but writable only by root:
pages_to_scan - how many present pages to scan before ksmd goes to sleep pages_to_scan
e.g. "echo 100 > /sys/kernel/mm/ksm/pages_to_scan" how many present pages to scan before ksmd goes to sleep
Default: 100 (chosen for demonstration purposes) e.g. ``echo 100 > /sys/kernel/mm/ksm/pages_to_scan`` Default: 100
(chosen for demonstration purposes)
sleep_millisecs - how many milliseconds ksmd should sleep before next scan
e.g. "echo 20 > /sys/kernel/mm/ksm/sleep_millisecs" sleep_millisecs
Default: 20 (chosen for demonstration purposes) how many milliseconds ksmd should sleep before next scan
e.g. ``echo 20 > /sys/kernel/mm/ksm/sleep_millisecs`` Default: 20
merge_across_nodes - specifies if pages from different numa nodes can be merged. (chosen for demonstration purposes)
When set to 0, ksm merges only pages which physically
reside in the memory area of same NUMA node. That brings merge_across_nodes
lower latency to access of shared pages. Systems with more specifies if pages from different numa nodes can be merged.
nodes, at significant NUMA distances, are likely to benefit When set to 0, ksm merges only pages which physically reside
from the lower latency of setting 0. Smaller systems, which in the memory area of same NUMA node. That brings lower
need to minimize memory usage, are likely to benefit from latency to access of shared pages. Systems with more nodes, at
the greater sharing of setting 1 (default). You may wish to significant NUMA distances, are likely to benefit from the
compare how your system performs under each setting, before lower latency of setting 0. Smaller systems, which need to
deciding on which to use. merge_across_nodes setting can be minimize memory usage, are likely to benefit from the greater
changed only when there are no ksm shared pages in system: sharing of setting 1 (default). You may wish to compare how
set run 2 to unmerge pages first, then to 1 after changing your system performs under each setting, before deciding on
merge_across_nodes, to remerge according to the new setting. which to use. merge_across_nodes setting can be changed only
Default: 1 (merging across nodes as in earlier releases) when there are no ksm shared pages in system: set run 2 to
unmerge pages first, then to 1 after changing
run - set 0 to stop ksmd from running but keep merged pages, merge_across_nodes, to remerge according to the new setting.
set 1 to run ksmd e.g. "echo 1 > /sys/kernel/mm/ksm/run", Default: 1 (merging across nodes as in earlier releases)
set 2 to stop ksmd and unmerge all pages currently merged,
but leave mergeable areas registered for next run run
Default: 0 (must be changed to 1 to activate KSM, set 0 to stop ksmd from running but keep merged pages,
except if CONFIG_SYSFS is disabled) set 1 to run ksmd e.g. ``echo 1 > /sys/kernel/mm/ksm/run``,
set 2 to stop ksmd and unmerge all pages currently merged, but
use_zero_pages - specifies whether empty pages (i.e. allocated pages leave mergeable areas registered for next run Default: 0 (must
that only contain zeroes) should be treated specially. be changed to 1 to activate KSM, except if CONFIG_SYSFS is
When set to 1, empty pages are merged with the kernel disabled)
zero page(s) instead of with each other as it would
happen normally. This can improve the performance on use_zero_pages
architectures with coloured zero pages, depending on specifies whether empty pages (i.e. allocated pages that only
the workload. Care should be taken when enabling this contain zeroes) should be treated specially. When set to 1,
setting, as it can potentially degrade the performance empty pages are merged with the kernel zero page(s) instead of
of KSM for some workloads, for example if the checksums with each other as it would happen normally. This can improve
of pages candidate for merging match the checksum of the performance on architectures with coloured zero pages,
an empty page. This setting can be changed at any time, depending on the workload. Care should be taken when enabling
it is only effective for pages merged after the change. this setting, as it can potentially degrade the performance of
Default: 0 (normal KSM behaviour as in earlier releases) KSM for some workloads, for example if the checksums of pages
candidate for merging match the checksum of an empty
max_page_sharing - Maximum sharing allowed for each KSM page. This page. This setting can be changed at any time, it is only
enforces a deduplication limit to avoid the virtual effective for pages merged after the change. Default: 0
memory rmap lists to grow too large. The minimum (normal KSM behaviour as in earlier releases)
value is 2 as a newly created KSM page will have at
least two sharers. The rmap walk has O(N) max_page_sharing
complexity where N is the number of rmap_items Maximum sharing allowed for each KSM page. This enforces a
(i.e. virtual mappings) that are sharing the page, deduplication limit to avoid the virtual memory rmap lists to
which is in turn capped by max_page_sharing. So grow too large. The minimum value is 2 as a newly created KSM
this effectively spread the the linear O(N) page will have at least two sharers. The rmap walk has O(N)
computational complexity from rmap walk context complexity where N is the number of rmap_items (i.e. virtual
over different KSM pages. The ksmd walk over the mappings) that are sharing the page, which is in turn capped
stable_node "chains" is also O(N), but N is the by max_page_sharing. So this effectively spread the the linear
number of stable_node "dups", not the number of O(N) computational complexity from rmap walk context over
rmap_items, so it has not a significant impact on different KSM pages. The ksmd walk over the stable_node
ksmd performance. In practice the best stable_node "chains" is also O(N), but N is the number of stable_node
"dup" candidate will be kept and found at the head "dups", not the number of rmap_items, so it has not a
of the "dups" list. The higher this value the significant impact on ksmd performance. In practice the best
faster KSM will merge the memory (because there stable_node "dup" candidate will be kept and found at the head
will be fewer stable_node dups queued into the of the "dups" list. The higher this value the faster KSM will
stable_node chain->hlist to check for pruning) and merge the memory (because there will be fewer stable_node dups
the higher the deduplication factor will be, but queued into the stable_node chain->hlist to check for pruning)
the slowest the worst case rmap walk could be for and the higher the deduplication factor will be, but the
any given KSM page. Slowing down the rmap_walk slowest the worst case rmap walk could be for any given KSM
means there will be higher latency for certain page. Slowing down the rmap_walk means there will be higher
virtual memory operations happening during latency for certain virtual memory operations happening during
swapping, compaction, NUMA balancing and page swapping, compaction, NUMA balancing and page migration, in
migration, in turn decreasing responsiveness for turn decreasing responsiveness for the caller of those virtual
the caller of those virtual memory operations. The memory operations. The scheduler latency of other tasks not
scheduler latency of other tasks not involved with involved with the VM operations doing the rmap walk is not
the VM operations doing the rmap walk is not affected by this parameter as the rmap walks are always
affected by this parameter as the rmap walks are schedule friendly themselves.
always schedule friendly themselves.
stable_node_chains_prune_millisecs
stable_node_chains_prune_millisecs - How frequently to walk the whole How frequently to walk the whole list of stable_node "dups"
list of stable_node "dups" linked in the linked in the stable_node "chains" in order to prune stale
stable_node "chains" in order to prune stale stable_nodes. Smaller milllisecs values will free up the KSM
stable_nodes. Smaller milllisecs values will free metadata with lower latency, but they will make ksmd use more
up the KSM metadata with lower latency, but they CPU during the scan. This only applies to the stable_node
will make ksmd use more CPU during the scan. This chains so it's a noop if not a single KSM page hit the
only applies to the stable_node chains so it's a max_page_sharing yet (there would be no stable_node chains in
noop if not a single KSM page hit the such case).
max_page_sharing yet (there would be no stable_node
chains in such case). The effectiveness of KSM and MADV_MERGEABLE is shown in ``/sys/kernel/mm/ksm/``:
The effectiveness of KSM and MADV_MERGEABLE is shown in /sys/kernel/mm/ksm/: pages_shared
how many shared pages are being used
pages_shared - how many shared pages are being used pages_sharing
pages_sharing - how many more sites are sharing them i.e. how much saved how many more sites are sharing them i.e. how much saved
pages_unshared - how many pages unique but repeatedly checked for merging pages_unshared
pages_volatile - how many pages changing too fast to be placed in a tree how many pages unique but repeatedly checked for merging
full_scans - how many times all mergeable areas have been scanned pages_volatile
how many pages changing too fast to be placed in a tree
stable_node_chains - number of stable node chains allocated, this is full_scans
effectively the number of KSM pages that hit the how many times all mergeable areas have been scanned
max_page_sharing limit stable_node_chains
stable_node_dups - number of stable node dups queued into the number of stable node chains allocated, this is effectively
stable_node chains the number of KSM pages that hit the max_page_sharing limit
stable_node_dups
number of stable node dups queued into the stable_node chains
A high ratio of pages_sharing to pages_shared indicates good sharing, but A high ratio of pages_sharing to pages_shared indicates good sharing, but
a high ratio of pages_unshared to pages_sharing indicates wasted effort. a high ratio of pages_unshared to pages_sharing indicates wasted effort.
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7