1How to use the Kernel Samepage Merging feature 2---------------------------------------------- 3 4KSM is a memory-saving de-duplication feature, enabled by CONFIG_KSM=y, 5added to the Linux kernel in 2.6.32. See mm/ksm.c for its implementation, 6and http://lwn.net/Articles/306704/ and http://lwn.net/Articles/330589/ 7 8The KSM daemon ksmd periodically scans those areas of user memory which 9have been registered with it, looking for pages of identical content which 10can be replaced by a single write-protected page (which is automatically 11copied if a process later wants to update its content). 12 13KSM was originally developed for use with KVM (where it was known as 14Kernel Shared Memory), to fit more virtual machines into physical memory, 15by sharing the data common between them. But it can be useful to any 16application which generates many instances of the same data. 17 18KSM only merges anonymous (private) pages, never pagecache (file) pages. 19KSM's merged pages were originally locked into kernel memory, but can now 20be swapped out just like other user pages (but sharing is broken when they 21are swapped back in: ksmd must rediscover their identity and merge again). 22 23KSM only operates on those areas of address space which an application 24has advised to be likely candidates for merging, by using the madvise(2) 25system call: int madvise(addr, length, MADV_MERGEABLE). 26 27The app may call int madvise(addr, length, MADV_UNMERGEABLE) to cancel 28that advice and restore unshared pages: whereupon KSM unmerges whatever 29it merged in that range. Note: this unmerging call may suddenly require 30more memory than is available - possibly failing with EAGAIN, but more 31probably arousing the Out-Of-Memory killer. 32 33If KSM is not configured into the running kernel, madvise MADV_MERGEABLE 34and MADV_UNMERGEABLE simply fail with EINVAL. If the running kernel was 35built with CONFIG_KSM=y, those calls will normally succeed: even if the 36the KSM daemon is not currently running, MADV_MERGEABLE still registers 37the range for whenever the KSM daemon is started; even if the range 38cannot contain any pages which KSM could actually merge; even if 39MADV_UNMERGEABLE is applied to a range which was never MADV_MERGEABLE. 40 41Like other madvise calls, they are intended for use on mapped areas of 42the user address space: they will report ENOMEM if the specified range 43includes unmapped gaps (though working on the intervening mapped areas), 44and might fail with EAGAIN if not enough memory for internal structures. 45 46Applications should be considerate in their use of MADV_MERGEABLE, 47restricting its use to areas likely to benefit. KSM's scans may use a lot 48of processing power: some installations will disable KSM for that reason. 49 50The KSM daemon is controlled by sysfs files in /sys/kernel/mm/ksm/, 51readable by all but writable only by root: 52 53pages_to_scan - how many present pages to scan before ksmd goes to sleep 54 e.g. "echo 100 > /sys/kernel/mm/ksm/pages_to_scan" 55 Default: 100 (chosen for demonstration purposes) 56 57sleep_millisecs - how many milliseconds ksmd should sleep before next scan 58 e.g. "echo 20 > /sys/kernel/mm/ksm/sleep_millisecs" 59 Default: 20 (chosen for demonstration purposes) 60 61run - set 0 to stop ksmd from running but keep merged pages, 62 set 1 to run ksmd e.g. "echo 1 > /sys/kernel/mm/ksm/run", 63 set 2 to stop ksmd and unmerge all pages currently merged, 64 but leave mergeable areas registered for next run 65 Default: 0 (must be changed to 1 to activate KSM, 66 except if CONFIG_SYSFS is disabled) 67 68The effectiveness of KSM and MADV_MERGEABLE is shown in /sys/kernel/mm/ksm/: 69 70pages_shared - how many shared pages are being used 71pages_sharing - how many more sites are sharing them i.e. how much saved 72pages_unshared - how many pages unique but repeatedly checked for merging 73pages_volatile - how many pages changing too fast to be placed in a tree 74full_scans - how many times all mergeable areas have been scanned 75 76A high ratio of pages_sharing to pages_shared indicates good sharing, but 77a high ratio of pages_unshared to pages_sharing indicates wasted effort. 78pages_volatile embraces several different kinds of activity, but a high 79proportion there would also indicate poor use of madvise MADV_MERGEABLE. 80 81Izik Eidus, 82Hugh Dickins, 17 Nov 2009 83