Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 1 | Overview: |
| 2 | |
| 3 | Zswap is a lightweight compressed cache for swap pages. It takes pages that are |
| 4 | in the process of being swapped out and attempts to compress them into a |
| 5 | dynamically allocated RAM-based memory pool. zswap basically trades CPU cycles |
| 6 | for potentially reduced swap I/O. This trade-off can also result in a |
| 7 | significant performance improvement if reads from the compressed cache are |
| 8 | faster than reads from a swap device. |
| 9 | |
| 10 | NOTE: Zswap is a new feature as of v3.11 and interacts heavily with memory |
Christian Hesse | 0151e3d | 2013-11-12 15:07:34 -0800 | [diff] [blame] | 11 | reclaim. This interaction has not been fully explored on the large set of |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 12 | potential configurations and workloads that exist. For this reason, zswap |
| 13 | is a work in progress and should be considered experimental. |
| 14 | |
| 15 | Some potential benefits: |
| 16 | * Desktop/laptop users with limited RAM capacities can mitigate the |
| 17 | performance impact of swapping. |
| 18 | * Overcommitted guests that share a common I/O resource can |
| 19 | dramatically reduce their swap I/O pressure, avoiding heavy handed I/O |
| 20 | throttling by the hypervisor. This allows more work to get done with less |
| 21 | impact to the guest workload and guests sharing the I/O subsystem |
| 22 | * Users with SSDs as swap devices can extend the life of the device by |
| 23 | drastically reducing life-shortening writes. |
| 24 | |
| 25 | Zswap evicts pages from compressed cache on an LRU basis to the backing swap |
Christian Hesse | 0151e3d | 2013-11-12 15:07:34 -0800 | [diff] [blame] | 26 | device when the compressed pool reaches its size limit. This requirement had |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 27 | been identified in prior community discussions. |
| 28 | |
Dan Streetman | c00ed16 | 2015-06-25 15:00:35 -0700 | [diff] [blame] | 29 | Zswap is disabled by default but can be enabled at boot time by setting |
| 30 | the "enabled" attribute to 1 at boot time. ie: zswap.enabled=1. Zswap |
| 31 | can also be enabled and disabled at runtime using the sysfs interface. |
| 32 | An example command to enable zswap at runtime, assuming sysfs is mounted |
| 33 | at /sys, is: |
| 34 | |
Dan Streetman | 9c4c5ef | 2015-09-09 15:35:25 -0700 | [diff] [blame] | 35 | echo 1 > /sys/module/zswap/parameters/enabled |
Dan Streetman | c00ed16 | 2015-06-25 15:00:35 -0700 | [diff] [blame] | 36 | |
| 37 | When zswap is disabled at runtime it will stop storing pages that are |
| 38 | being swapped out. However, it will _not_ immediately write out or fault |
| 39 | back into memory all of the pages stored in the compressed pool. The |
| 40 | pages stored in zswap will remain in the compressed pool until they are |
| 41 | either invalidated or faulted back into memory. In order to force all |
| 42 | pages out of the compressed pool, a swapoff on the swap device(s) will |
| 43 | fault back into memory all swapped out pages, including those in the |
| 44 | compressed pool. |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 45 | |
| 46 | Design: |
| 47 | |
| 48 | Zswap receives pages for compression through the Frontswap API and is able to |
| 49 | evict pages from its own compressed pool on an LRU basis and write them back to |
| 50 | the backing swap device in the case that the compressed pool is full. |
| 51 | |
Dan Streetman | 9c4c5ef | 2015-09-09 15:35:25 -0700 | [diff] [blame] | 52 | Zswap makes use of zpool for the managing the compressed memory pool. Each |
| 53 | allocation in zpool is not directly accessible by address. Rather, a handle is |
Christian Hesse | 0151e3d | 2013-11-12 15:07:34 -0800 | [diff] [blame] | 54 | returned by the allocation routine and that handle must be mapped before being |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 55 | accessed. The compressed memory pool grows on demand and shrinks as compressed |
Dan Streetman | 9c4c5ef | 2015-09-09 15:35:25 -0700 | [diff] [blame] | 56 | pages are freed. The pool is not preallocated. By default, a zpool of type |
| 57 | zbud is created, but it can be selected at boot time by setting the "zpool" |
| 58 | attribute, e.g. zswap.zpool=zbud. It can also be changed at runtime using the |
| 59 | sysfs "zpool" attribute, e.g. |
| 60 | |
| 61 | echo zbud > /sys/module/zswap/parameters/zpool |
| 62 | |
| 63 | The zbud type zpool allocates exactly 1 page to store 2 compressed pages, which |
| 64 | means the compression ratio will always be 2:1 or worse (because of half-full |
| 65 | zbud pages). The zsmalloc type zpool has a more complex compressed page |
| 66 | storage method, and it can achieve greater storage densities. However, |
| 67 | zsmalloc does not implement compressed page eviction, so once zswap fills it |
| 68 | cannot evict the oldest page, it can only reject new pages. |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 69 | |
| 70 | When a swap page is passed from frontswap to zswap, zswap maintains a mapping |
Dan Streetman | 9c4c5ef | 2015-09-09 15:35:25 -0700 | [diff] [blame] | 71 | of the swap entry, a combination of the swap type and swap offset, to the zpool |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 72 | handle that references that compressed swap page. This mapping is achieved |
| 73 | with a red-black tree per swap type. The swap offset is the search key for the |
| 74 | tree nodes. |
| 75 | |
| 76 | During a page fault on a PTE that is a swap entry, frontswap calls the zswap |
| 77 | load function to decompress the page into the page allocated by the page fault |
| 78 | handler. |
| 79 | |
| 80 | Once there are no PTEs referencing a swap page stored in zswap (i.e. the count |
| 81 | in the swap_map goes to 0) the swap code calls the zswap invalidate function, |
| 82 | via frontswap, to free the compressed entry. |
| 83 | |
| 84 | Zswap seeks to be simple in its policies. Sysfs attributes allow for one user |
Christian Hesse | 0151e3d | 2013-11-12 15:07:34 -0800 | [diff] [blame] | 85 | controlled policy: |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 86 | * max_pool_percent - The maximum percentage of memory that the compressed |
| 87 | pool can occupy. |
| 88 | |
Dan Streetman | 9c4c5ef | 2015-09-09 15:35:25 -0700 | [diff] [blame] | 89 | The default compressor is lzo, but it can be selected at boot time by setting |
| 90 | the “compressor” attribute, e.g. zswap.compressor=lzo. It can also be changed |
| 91 | at runtime using the sysfs "compressor" attribute, e.g. |
| 92 | |
| 93 | echo lzo > /sys/module/zswap/parameters/compressor |
| 94 | |
| 95 | When the zpool and/or compressor parameter is changed at runtime, any existing |
| 96 | compressed pages are not modified; they are left in their own zpool. When a |
| 97 | request is made for a page in an old zpool, it is uncompressed using its |
| 98 | original compressor. Once all pages are removed from an old zpool, the zpool |
| 99 | and its compressor are freed. |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 100 | |
Srividya Desireddy | 51f73ff | 2017-12-14 15:32:45 -0800 | [diff] [blame] | 101 | Some of the pages in zswap are same-value filled pages (i.e. contents of the |
| 102 | page have same value or repetitive pattern). These pages include zero-filled |
| 103 | pages and they are handled differently. During store operation, a page is |
| 104 | checked if it is a same-value filled page before compressing it. If true, the |
| 105 | compressed length of the page is set to zero and the pattern or same-filled |
| 106 | value is stored. |
| 107 | |
| 108 | Same-value filled pages identification feature is enabled by default and can be |
| 109 | disabled at boot time by setting the "same_filled_pages_enabled" attribute to 0, |
| 110 | e.g. zswap.same_filled_pages_enabled=0. It can also be enabled and disabled at |
| 111 | runtime using the sysfs "same_filled_pages_enabled" attribute, e.g. |
| 112 | |
| 113 | echo 1 > /sys/module/zswap/parameters/same_filled_pages_enabled |
| 114 | |
| 115 | When zswap same-filled page identification is disabled at runtime, it will stop |
| 116 | checking for the same-value filled pages during store operation. However, the |
| 117 | existing pages which are marked as same-value filled pages remain stored |
| 118 | unchanged in zswap until they are either loaded or invalidated. |
| 119 | |
Seth Jennings | 61b0d76 | 2013-07-10 16:05:05 -0700 | [diff] [blame] | 120 | A debugfs interface is provided for various statistic about pool size, number |
Srividya Desireddy | 51f73ff | 2017-12-14 15:32:45 -0800 | [diff] [blame] | 121 | of pages stored, same-value filled pages and various counters for the reasons |
| 122 | pages are rejected. |