Date: Sat, 6 Dec 1997 08:31:18 -0500 (EST) From: "John S. Dyson" <dyson@freebsd.org> To: current@freebsd.org Cc: dg@root.com Subject: VM system info Message-ID: <199712061331.IAA03947@dyson.iquest.net>
next in thread | raw e-mail | index | archive | help
Notes on VM tunables: I have recently added some interesting VM tunables. Since it would be nice if people other than me (or those who requested them) could use the features, I thought it would be nice to pass this info on: (My machine has 112MB, and I have modified a few of these, so if you get different results, don't worry :-)). vm.v_free_min: 287 A process waiting on memory will be waken up if: (vm cache memory + vm free memory) > vm.v_free_min vm.v_free_target: 1011 The pageout daemon tries to arrange that the total amount of free memory reaches this value. Note that the cache memory is included in the calculation of free memory. vm.v_free_reserved: 150 Processes will wakeup the pageout daemon and wait for memory if the number of free and cached pages drops below this value. vm.v_inactive_target: 5131 The pageout daemon tries to reach this many inactive pages when it runs. Inactive pages are often (but not always) removed from the process address space, and are also the pages that are cleaned. These pages can be quickly re-inserted into the process address space when a page fault happens. vm.v_cache_min: 3259 If the cache queue size drops very far below this value, then the pageout daemon is waken up. These pages can be quickly re-faulted back to where they were, or they can be used as free pages. The notion of "free pages" is mostly an anachronism on FreeBSD, but it would be best estimated as the number of cache pages plus the number of free pages. vm.v_pageout_free_min: 34 If the amount of free+cached memory drops below this value, then the pageout daemon goes into a memory conserving mode, thereby avoiding deadlock. vm.pageout_algorithm: 0 If this is "0" the pageout daemon uses our statistics based page management scheme, it it is "1" it uses a near LRU scheme. vm.swap_enabled: 1 Enables process swapping. vm.swap_idle_threshold1: 2 This is the guaranteed memory time for a process that is targeted for swapout because of severe low memory conditions. vm.swap_idle_threshold2: 10 This is the guaranteed memory time for a process that is targeted for swapout because of being idle. vm.pageout_stats_max: 1011 Maximum number of pages to scan for a stats run. (This should be close to v_free_target.) vm.pageout_full_stats_interval: 16 Interval in seconds for a full paging stats run. vm.pageout_stats_interval: 4 Interval in seconds for a limited paging stats run. vm.pageout_stats_free_max: 25 Maximum number of pages to operate on during a stats run. vm.swap_idle_enabled: 1 Enable the swap on idle capability. This allows supporting of prejudiced swapping of idle processes. It is very useful on timesharing systems. vm.defer_swapspace_pageouts: 0 Make the swap pager very conservative about swap space usage. (This is useful on systems that don't have enough swap space, and will let the system run longer without processes being killed.) There is a significant performance hit when this option is used, when the system is under memory pressure. vm.disable_swapspace_pageouts: 0 Almost totally disable pageouts. Unless an explicit msync is done (and normally the code doesn't support msync of swapspace anyway), the system will not use swap space -- with all of the associated advantages and disadvantages. vm.max_page_launder: 96 Maximum number of pages to clean per pass of the pageout daemon. (Currently, on my system, the value chosen by the pageout daemon is a little on the small side, but is a good compromise.) If you have a slow swap device, you might want to decrease this number, if you have several interleved swap devices, you might want to increase this number. -- John dyson@freebsd.org jdyson@nc.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199712061331.IAA03947>