1======================= 2Kernel Samepage Merging 3======================= 4 5Overview 6======== 7 8KSM is a memory-saving de-duplication feature, enabled by CONFIG_KSM=y, 9added to the Linux kernel in 2.6.32. See ``mm/ksm.c`` for its implementation, 10and http://lwn.net/Articles/306704/ and https://lwn.net/Articles/330589/ 11 12KSM was originally developed for use with KVM (where it was known as 13Kernel Shared Memory), to fit more virtual machines into physical memory, 14by sharing the data common between them. But it can be useful to any 15application which generates many instances of the same data. 16 17The KSM daemon ksmd periodically scans those areas of user memory 18which have been registered with it, looking for pages of identical 19content which can be replaced by a single write-protected page (which 20is automatically copied if a process later wants to update its 21content). The amount of pages that KSM daemon scans in a single pass 22and the time between the passes are configured using :ref:`sysfs 23interface <ksm_sysfs>` 24 25KSM only merges anonymous (private) pages, never pagecache (file) pages. 26KSM's merged pages were originally locked into kernel memory, but can now 27be swapped out just like other user pages (but sharing is broken when they 28are swapped back in: ksmd must rediscover their identity and merge again). 29 30Controlling KSM with madvise 31============================ 32 33KSM only operates on those areas of address space which an application 34has advised to be likely candidates for merging, by using the madvise(2) 35system call:: 36 37 int madvise(addr, length, MADV_MERGEABLE) 38 39The app may call 40 41:: 42 43 int madvise(addr, length, MADV_UNMERGEABLE) 44 45to cancel that advice and restore unshared pages: whereupon KSM 46unmerges whatever it merged in that range. Note: this unmerging call 47may suddenly require more memory than is available - possibly failing 48with EAGAIN, but more probably arousing the Out-Of-Memory killer. 49 50If KSM is not configured into the running kernel, madvise MADV_MERGEABLE 51and MADV_UNMERGEABLE simply fail with EINVAL. If the running kernel was 52built with CONFIG_KSM=y, those calls will normally succeed: even if the 53KSM daemon is not currently running, MADV_MERGEABLE still registers 54the range for whenever the KSM daemon is started; even if the range 55cannot contain any pages which KSM could actually merge; even if 56MADV_UNMERGEABLE is applied to a range which was never MADV_MERGEABLE. 57 58If a region of memory must be split into at least one new MADV_MERGEABLE 59or MADV_UNMERGEABLE region, the madvise may return ENOMEM if the process 60will exceed ``vm.max_map_count`` (see Documentation/admin-guide/sysctl/vm.rst). 61 62Like other madvise calls, they are intended for use on mapped areas of 63the user address space: they will report ENOMEM if the specified range 64includes unmapped gaps (though working on the intervening mapped areas), 65and might fail with EAGAIN if not enough memory for internal structures. 66 67Applications should be considerate in their use of MADV_MERGEABLE, 68restricting its use to areas likely to benefit. KSM's scans may use a lot 69of processing power: some installations will disable KSM for that reason. 70 71.. _ksm_sysfs: 72 73KSM daemon sysfs interface 74========================== 75 76The KSM daemon is controlled by sysfs files in ``/sys/kernel/mm/ksm/``, 77readable by all but writable only by root: 78 79pages_to_scan 80 how many pages to scan before ksmd goes to sleep 81 e.g. ``echo 100 > /sys/kernel/mm/ksm/pages_to_scan``. 82 83 Default: 100 (chosen for demonstration purposes) 84 85sleep_millisecs 86 how many milliseconds ksmd should sleep before next scan 87 e.g. ``echo 20 > /sys/kernel/mm/ksm/sleep_millisecs`` 88 89 Default: 20 (chosen for demonstration purposes) 90 91merge_across_nodes 92 specifies if pages from different NUMA nodes can be merged. 93 When set to 0, ksm merges only pages which physically reside 94 in the memory area of same NUMA node. That brings lower 95 latency to access of shared pages. Systems with more nodes, at 96 significant NUMA distances, are likely to benefit from the 97 lower latency of setting 0. Smaller systems, which need to 98 minimize memory usage, are likely to benefit from the greater 99 sharing of setting 1 (default). You may wish to compare how 100 your system performs under each setting, before deciding on 101 which to use. ``merge_across_nodes`` setting can be changed only 102 when there are no ksm shared pages in the system: set run 2 to 103 unmerge pages first, then to 1 after changing 104 ``merge_across_nodes``, to remerge according to the new setting. 105 106 Default: 1 (merging across nodes as in earlier releases) 107 108run 109 * set to 0 to stop ksmd from running but keep merged pages, 110 * set to 1 to run ksmd e.g. ``echo 1 > /sys/kernel/mm/ksm/run``, 111 * set to 2 to stop ksmd and unmerge all pages currently merged, but 112 leave mergeable areas registered for next run. 113 114 Default: 0 (must be changed to 1 to activate KSM, except if 115 CONFIG_SYSFS is disabled) 116 117use_zero_pages 118 specifies whether empty pages (i.e. allocated pages that only 119 contain zeroes) should be treated specially. When set to 1, 120 empty pages are merged with the kernel zero page(s) instead of 121 with each other as it would happen normally. This can improve 122 the performance on architectures with coloured zero pages, 123 depending on the workload. Care should be taken when enabling 124 this setting, as it can potentially degrade the performance of 125 KSM for some workloads, for example if the checksums of pages 126 candidate for merging match the checksum of an empty 127 page. This setting can be changed at any time, it is only 128 effective for pages merged after the change. 129 130 Default: 0 (normal KSM behaviour as in earlier releases) 131 132max_page_sharing 133 Maximum sharing allowed for each KSM page. This enforces a 134 deduplication limit to avoid high latency for virtual memory 135 operations that involve traversal of the virtual mappings that 136 share the KSM page. The minimum value is 2 as a newly created 137 KSM page will have at least two sharers. The higher this value 138 the faster KSM will merge the memory and the higher the 139 deduplication factor will be, but the slower the worst case 140 virtual mappings traversal could be for any given KSM 141 page. Slowing down this traversal means there will be higher 142 latency for certain virtual memory operations happening during 143 swapping, compaction, NUMA balancing and page migration, in 144 turn decreasing responsiveness for the caller of those virtual 145 memory operations. The scheduler latency of other tasks not 146 involved with the VM operations doing the virtual mappings 147 traversal is not affected by this parameter as these 148 traversals are always schedule friendly themselves. 149 150stable_node_chains_prune_millisecs 151 specifies how frequently KSM checks the metadata of the pages 152 that hit the deduplication limit for stale information. 153 Smaller milllisecs values will free up the KSM metadata with 154 lower latency, but they will make ksmd use more CPU during the 155 scan. It's a noop if not a single KSM page hit the 156 ``max_page_sharing`` yet. 157 158The effectiveness of KSM and MADV_MERGEABLE is shown in ``/sys/kernel/mm/ksm/``: 159 160general_profit 161 how effective is KSM. The calculation is explained below. 162pages_scanned 163 how many pages are being scanned for ksm 164pages_shared 165 how many shared pages are being used 166pages_sharing 167 how many more sites are sharing them i.e. how much saved 168pages_unshared 169 how many pages unique but repeatedly checked for merging 170pages_volatile 171 how many pages changing too fast to be placed in a tree 172full_scans 173 how many times all mergeable areas have been scanned 174stable_node_chains 175 the number of KSM pages that hit the ``max_page_sharing`` limit 176stable_node_dups 177 number of duplicated KSM pages 178ksm_zero_pages 179 how many zero pages that are still mapped into processes were mapped by 180 KSM when deduplicating. 181 182When ``use_zero_pages`` is/was enabled, the sum of ``pages_sharing`` + 183``ksm_zero_pages`` represents the actual number of pages saved by KSM. 184if ``use_zero_pages`` has never been enabled, ``ksm_zero_pages`` is 0. 185 186A high ratio of ``pages_sharing`` to ``pages_shared`` indicates good 187sharing, but a high ratio of ``pages_unshared`` to ``pages_sharing`` 188indicates wasted effort. ``pages_volatile`` embraces several 189different kinds of activity, but a high proportion there would also 190indicate poor use of madvise MADV_MERGEABLE. 191 192The maximum possible ``pages_sharing/pages_shared`` ratio is limited by the 193``max_page_sharing`` tunable. To increase the ratio ``max_page_sharing`` must 194be increased accordingly. 195 196Monitoring KSM profit 197===================== 198 199KSM can save memory by merging identical pages, but also can consume 200additional memory, because it needs to generate a number of rmap_items to 201save each scanned page's brief rmap information. Some of these pages may 202be merged, but some may not be abled to be merged after being checked 203several times, which are unprofitable memory consumed. 204 2051) How to determine whether KSM save memory or consume memory in system-wide 206 range? Here is a simple approximate calculation for reference:: 207 208 general_profit =~ ksm_saved_pages * sizeof(page) - (all_rmap_items) * 209 sizeof(rmap_item); 210 211 where ksm_saved_pages equals to the sum of ``pages_sharing`` + 212 ``ksm_zero_pages`` of the system, and all_rmap_items can be easily 213 obtained by summing ``pages_sharing``, ``pages_shared``, ``pages_unshared`` 214 and ``pages_volatile``. 215 2162) The KSM profit inner a single process can be similarly obtained by the 217 following approximate calculation:: 218 219 process_profit =~ ksm_saved_pages * sizeof(page) - 220 ksm_rmap_items * sizeof(rmap_item). 221 222 where ksm_saved_pages equals to the sum of ``ksm_merging_pages`` and 223 ``ksm_zero_pages``, both of which are shown under the directory 224 ``/proc/<pid>/ksm_stat``, and ksm_rmap_items is also shown in 225 ``/proc/<pid>/ksm_stat``. The process profit is also shown in 226 ``/proc/<pid>/ksm_stat`` as ksm_process_profit. 227 228From the perspective of application, a high ratio of ``ksm_rmap_items`` to 229``ksm_merging_pages`` means a bad madvise-applied policy, so developers or 230administrators have to rethink how to change madvise policy. Giving an example 231for reference, a page's size is usually 4K, and the rmap_item's size is 232separately 32B on 32-bit CPU architecture and 64B on 64-bit CPU architecture. 233so if the ``ksm_rmap_items/ksm_merging_pages`` ratio exceeds 64 on 64-bit CPU 234or exceeds 128 on 32-bit CPU, then the app's madvise policy should be dropped, 235because the ksm profit is approximately zero or negative. 236 237Monitoring KSM events 238===================== 239 240There are some counters in /proc/vmstat that may be used to monitor KSM events. 241KSM might help save memory, it's a tradeoff by may suffering delay on KSM COW 242or on swapping in copy. Those events could help users evaluate whether or how 243to use KSM. For example, if cow_ksm increases too fast, user may decrease the 244range of madvise(, , MADV_MERGEABLE). 245 246cow_ksm 247 is incremented every time a KSM page triggers copy on write (COW) 248 when users try to write to a KSM page, we have to make a copy. 249 250ksm_swpin_copy 251 is incremented every time a KSM page is copied when swapping in 252 note that KSM page might be copied when swapping in because do_swap_page() 253 cannot do all the locking needed to reconstitute a cross-anon_vma KSM page. 254 255-- 256Izik Eidus, 257Hugh Dickins, 17 Nov 2009 258