1 =======================
2 Kernel Samepage Merging
3 =======================
8 KSM is a memory-saving de-duplication feature, enabled by CONFIG_KSM=y,
9 added to the Linux kernel in 2.6.32. See ``mm/ksm.c`` for its implementation,
10 and http://lwn.net/Articles/306704/ and https://lwn.net/Articles/330589/
12 KSM was originally developed for use with KVM (where it was known as
13 Kernel Shared Memory), to fit more virtual machines into physical memory,
14 by sharing the data common between them. But it can be useful to any
15 application which generates many instances of the same data.
17 The KSM daemon ksmd periodically scans those areas of user memory
18 which have been registered with it, looking for pages of identical
19 content which can be replaced by a single write-protected page (which
20 is automatically copied if a process later wants to update its
21 content). The amount of pages that KSM daemon scans in a single pass
22 and the time between the passes are configured using :ref:`sysfs
23 interface <ksm_sysfs>`
25 KSM only merges anonymous (private) pages, never pagecache (file) pages.
26 KSM's merged pages were originally locked into kernel memory, but can now
27 be swapped out just like other user pages (but sharing is broken when they
28 are swapped back in: ksmd must rediscover their identity and merge again).
30 Controlling KSM with madvise
31 ============================
33 KSM only operates on those areas of address space which an application
34 has advised to be likely candidates for merging, by using the madvise(2)
37 int madvise(addr, length, MADV_MERGEABLE)
43 int madvise(addr, length, MADV_UNMERGEABLE)
45 to cancel that advice and restore unshared pages: whereupon KSM
46 unmerges whatever it merged in that range. Note: this unmerging call
47 may suddenly require more memory than is available - possibly failing
48 with EAGAIN, but more probably arousing the Out-Of-Memory killer.
50 If KSM is not configured into the running kernel, madvise MADV_MERGEABLE
51 and MADV_UNMERGEABLE simply fail with EINVAL. If the running kernel was
52 built with CONFIG_KSM=y, those calls will normally succeed: even if the
53 KSM daemon is not currently running, MADV_MERGEABLE still registers
54 the range for whenever the KSM daemon is started; even if the range
55 cannot contain any pages which KSM could actually merge; even if
56 MADV_UNMERGEABLE is applied to a range which was never MADV_MERGEABLE.
58 If a region of memory must be split into at least one new MADV_MERGEABLE
59 or MADV_UNMERGEABLE region, the madvise may return ENOMEM if the process
60 will exceed ``vm.max_map_count`` (see Documentation/admin-guide/sysctl/vm.rst).
62 Like other madvise calls, they are intended for use on mapped areas of
63 the user address space: they will report ENOMEM if the specified range
64 includes unmapped gaps (though working on the intervening mapped areas),
65 and might fail with EAGAIN if not enough memory for internal structures.
67 Applications should be considerate in their use of MADV_MERGEABLE,
68 restricting its use to areas likely to benefit. KSM's scans may use a lot
69 of processing power: some installations will disable KSM for that reason.
73 KSM daemon sysfs interface
74 ==========================
76 The KSM daemon is controlled by sysfs files in ``/sys/kernel/mm/ksm/``,
77 readable by all but writable only by root:
80 how many pages to scan before ksmd goes to sleep
81 e.g. ``echo 100 > /sys/kernel/mm/ksm/pages_to_scan``.
83 The pages_to_scan value cannot be changed if ``advisor_mode`` has
84 been set to scan-time.
86 Default: 100 (chosen for demonstration purposes)
89 how many milliseconds ksmd should sleep before next scan
90 e.g. ``echo 20 > /sys/kernel/mm/ksm/sleep_millisecs``
92 Default: 20 (chosen for demonstration purposes)
95 specifies if pages from different NUMA nodes can be merged.
96 When set to 0, ksm merges only pages which physically reside
97 in the memory area of same NUMA node. That brings lower
98 latency to access of shared pages. Systems with more nodes, at
99 significant NUMA distances, are likely to benefit from the
100 lower latency of setting 0. Smaller systems, which need to
101 minimize memory usage, are likely to benefit from the greater
102 sharing of setting 1 (default). You may wish to compare how
103 your system performs under each setting, before deciding on
104 which to use. ``merge_across_nodes`` setting can be changed only
105 when there are no ksm shared pages in the system: set run 2 to
106 unmerge pages first, then to 1 after changing
107 ``merge_across_nodes``, to remerge according to the new setting.
109 Default: 1 (merging across nodes as in earlier releases)
112 * set to 0 to stop ksmd from running but keep merged pages,
113 * set to 1 to run ksmd e.g. ``echo 1 > /sys/kernel/mm/ksm/run``,
114 * set to 2 to stop ksmd and unmerge all pages currently merged, but
115 leave mergeable areas registered for next run.
117 Default: 0 (must be changed to 1 to activate KSM, except if
118 CONFIG_SYSFS is disabled)
121 specifies whether empty pages (i.e. allocated pages that only
122 contain zeroes) should be treated specially. When set to 1,
123 empty pages are merged with the kernel zero page(s) instead of
124 with each other as it would happen normally. This can improve
125 the performance on architectures with coloured zero pages,
126 depending on the workload. Care should be taken when enabling
127 this setting, as it can potentially degrade the performance of
128 KSM for some workloads, for example if the checksums of pages
129 candidate for merging match the checksum of an empty
130 page. This setting can be changed at any time, it is only
131 effective for pages merged after the change.
133 Default: 0 (normal KSM behaviour as in earlier releases)
136 Maximum sharing allowed for each KSM page. This enforces a
137 deduplication limit to avoid high latency for virtual memory
138 operations that involve traversal of the virtual mappings that
139 share the KSM page. The minimum value is 2 as a newly created
140 KSM page will have at least two sharers. The higher this value
141 the faster KSM will merge the memory and the higher the
142 deduplication factor will be, but the slower the worst case
143 virtual mappings traversal could be for any given KSM
144 page. Slowing down this traversal means there will be higher
145 latency for certain virtual memory operations happening during
146 swapping, compaction, NUMA balancing and page migration, in
147 turn decreasing responsiveness for the caller of those virtual
148 memory operations. The scheduler latency of other tasks not
149 involved with the VM operations doing the virtual mappings
150 traversal is not affected by this parameter as these
151 traversals are always schedule friendly themselves.
153 stable_node_chains_prune_millisecs
154 specifies how frequently KSM checks the metadata of the pages
155 that hit the deduplication limit for stale information.
156 Smaller milllisecs values will free up the KSM metadata with
157 lower latency, but they will make ksmd use more CPU during the
158 scan. It's a noop if not a single KSM page hit the
159 ``max_page_sharing`` yet.
162 Historically KSM checked every candidate page for each scan. It did
163 not take into account historic information. When smart scan is
164 enabled, pages that have previously not been de-duplicated get
165 skipped. How often these pages are skipped depends on how often
166 de-duplication has already been tried and failed. By default this
167 optimization is enabled. The ``pages_skipped`` metric shows how
168 effective the setting is.
171 The ``advisor_mode`` selects the current advisor. Two modes are
172 supported: none and scan-time. The default is none. By setting
173 ``advisor_mode`` to scan-time, the scan time advisor is enabled.
174 The section about ``advisor`` explains in detail how the scan time
178 specifies the upper limit of the cpu percent usage of the ksmd
179 background thread. The default is 70.
181 advisor_target_scan_time
182 specifies the target scan time in seconds to scan all the candidate
183 pages. The default value is 200 seconds.
185 advisor_min_pages_to_scan
186 specifies the lower limit of the ``pages_to_scan`` parameter of the
187 scan time advisor. The default is 500.
189 adivsor_max_pages_to_scan
190 specifies the upper limit of the ``pages_to_scan`` parameter of the
191 scan time advisor. The default is 30000.
193 The effectiveness of KSM and MADV_MERGEABLE is shown in ``/sys/kernel/mm/ksm/``:
196 how effective is KSM. The calculation is explained below.
198 how many pages are being scanned for ksm
200 how many shared pages are being used
202 how many more sites are sharing them i.e. how much saved
204 how many pages unique but repeatedly checked for merging
206 how many pages changing too fast to be placed in a tree
208 how many pages did the "smart" page scanning algorithm skip
210 how many times all mergeable areas have been scanned
212 the number of KSM pages that hit the ``max_page_sharing`` limit
214 number of duplicated KSM pages
216 how many zero pages that are still mapped into processes were mapped by
217 KSM when deduplicating.
219 When ``use_zero_pages`` is/was enabled, the sum of ``pages_sharing`` +
220 ``ksm_zero_pages`` represents the actual number of pages saved by KSM.
221 if ``use_zero_pages`` has never been enabled, ``ksm_zero_pages`` is 0.
223 A high ratio of ``pages_sharing`` to ``pages_shared`` indicates good
224 sharing, but a high ratio of ``pages_unshared`` to ``pages_sharing``
225 indicates wasted effort. ``pages_volatile`` embraces several
226 different kinds of activity, but a high proportion there would also
227 indicate poor use of madvise MADV_MERGEABLE.
229 The maximum possible ``pages_sharing/pages_shared`` ratio is limited by the
230 ``max_page_sharing`` tunable. To increase the ratio ``max_page_sharing`` must
231 be increased accordingly.
233 Monitoring KSM profit
234 =====================
236 KSM can save memory by merging identical pages, but also can consume
237 additional memory, because it needs to generate a number of rmap_items to
238 save each scanned page's brief rmap information. Some of these pages may
239 be merged, but some may not be abled to be merged after being checked
240 several times, which are unprofitable memory consumed.
242 1) How to determine whether KSM save memory or consume memory in system-wide
243 range? Here is a simple approximate calculation for reference::
245 general_profit =~ ksm_saved_pages * sizeof(page) - (all_rmap_items) *
248 where ksm_saved_pages equals to the sum of ``pages_sharing`` +
249 ``ksm_zero_pages`` of the system, and all_rmap_items can be easily
250 obtained by summing ``pages_sharing``, ``pages_shared``, ``pages_unshared``
251 and ``pages_volatile``.
253 2) The KSM profit inner a single process can be similarly obtained by the
254 following approximate calculation::
256 process_profit =~ ksm_saved_pages * sizeof(page) -
257 ksm_rmap_items * sizeof(rmap_item).
259 where ksm_saved_pages equals to the sum of ``ksm_merging_pages`` and
260 ``ksm_zero_pages``, both of which are shown under the directory
261 ``/proc/<pid>/ksm_stat``, and ksm_rmap_items is also shown in
262 ``/proc/<pid>/ksm_stat``. The process profit is also shown in
263 ``/proc/<pid>/ksm_stat`` as ksm_process_profit.
265 From the perspective of application, a high ratio of ``ksm_rmap_items`` to
266 ``ksm_merging_pages`` means a bad madvise-applied policy, so developers or
267 administrators have to rethink how to change madvise policy. Giving an example
268 for reference, a page's size is usually 4K, and the rmap_item's size is
269 separately 32B on 32-bit CPU architecture and 64B on 64-bit CPU architecture.
270 so if the ``ksm_rmap_items/ksm_merging_pages`` ratio exceeds 64 on 64-bit CPU
271 or exceeds 128 on 32-bit CPU, then the app's madvise policy should be dropped,
272 because the ksm profit is approximately zero or negative.
274 Monitoring KSM events
275 =====================
277 There are some counters in /proc/vmstat that may be used to monitor KSM events.
278 KSM might help save memory, it's a tradeoff by may suffering delay on KSM COW
279 or on swapping in copy. Those events could help users evaluate whether or how
280 to use KSM. For example, if cow_ksm increases too fast, user may decrease the
281 range of madvise(, , MADV_MERGEABLE).
284 is incremented every time a KSM page triggers copy on write (COW)
285 when users try to write to a KSM page, we have to make a copy.
288 is incremented every time a KSM page is copied when swapping in
289 note that KSM page might be copied when swapping in because do_swap_page()
290 cannot do all the locking needed to reconstitute a cross-anon_vma KSM page.
295 The number of candidate pages for KSM is dynamic. It can be often observed
296 that during the startup of an application more candidate pages need to be
297 processed. Without an advisor the ``pages_to_scan`` parameter needs to be
298 sized for the maximum number of candidate pages. The scan time advisor can
299 changes the ``pages_to_scan`` parameter based on demand.
301 The advisor can be enabled, so KSM can automatically adapt to changes in the
302 number of candidate pages to scan. Two advisors are implemented: none and
303 scan-time. With none, no advisor is enabled. The default is none.
305 The scan time advisor changes the ``pages_to_scan`` parameter based on the
306 observed scan times. The possible values for the ``pages_to_scan`` parameter is
307 limited by the ``advisor_max_cpu`` parameter. In addition there is also the
308 ``advisor_target_scan_time`` parameter. This parameter sets the target time to
309 scan all the KSM candidate pages. The parameter ``advisor_target_scan_time``
310 decides how aggressive the scan time advisor scans candidate pages. Lower
311 values make the scan time advisor to scan more aggressively. This is the most
312 important parameter for the configuration of the scan time advisor.
314 The initial value and the maximum value can be changed with
315 ``advisor_min_pages_to_scan`` and ``advisor_max_pages_to_scan``. The default
316 values are sufficient for most workloads and use cases.
318 The ``pages_to_scan`` parameter is re-calculated after a scan has been completed.
323 Hugh Dickins, 17 Nov 2009