Based on kernel version 4.3. Page generated on 2015-11-02 12:44 EST.
1 Guidance for writing policies 2 ============================= 3 4 Try to keep transactionality out of it. The core is careful to 5 avoid asking about anything that is migrating. This is a pain, but 6 makes it easier to write the policies. 7 8 Mappings are loaded into the policy at construction time. 9 10 Every bio that is mapped by the target is referred to the policy. 11 The policy can return a simple HIT or MISS or issue a migration. 12 13 Currently there's no way for the policy to issue background work, 14 e.g. to start writing back dirty blocks that are going to be evicte 15 soon. 16 17 Because we map bios, rather than requests it's easy for the policy 18 to get fooled by many small bios. For this reason the core target 19 issues periodic ticks to the policy. It's suggested that the policy 20 doesn't update states (eg, hit counts) for a block more than once 21 for each tick. The core ticks by watching bios complete, and so 22 trying to see when the io scheduler has let the ios run. 23 24 25 Overview of supplied cache replacement policies 26 =============================================== 27 28 multiqueue (mq) 29 --------------- 30 31 This policy has been deprecated in favor of the smq policy (see below). 32 33 The multiqueue policy has three sets of 16 queues: one set for entries 34 waiting for the cache and another two for those in the cache (a set for 35 clean entries and a set for dirty entries). 36 37 Cache entries in the queues are aged based on logical time. Entry into 38 the cache is based on variable thresholds and queue selection is based 39 on hit count on entry. The policy aims to take different cache miss 40 costs into account and to adjust to varying load patterns automatically. 41 42 Message and constructor argument pairs are: 43 'sequential_threshold <#nr_sequential_ios>' 44 'random_threshold <#nr_random_ios>' 45 'read_promote_adjustment <value>' 46 'write_promote_adjustment <value>' 47 'discard_promote_adjustment <value>' 48 49 The sequential threshold indicates the number of contiguous I/Os 50 required before a stream is treated as sequential. Once a stream is 51 considered sequential it will bypass the cache. The random threshold 52 is the number of intervening non-contiguous I/Os that must be seen 53 before the stream is treated as random again. 54 55 The sequential and random thresholds default to 512 and 4 respectively. 56 57 Large, sequential I/Os are probably better left on the origin device 58 since spindles tend to have good sequential I/O bandwidth. The 59 io_tracker counts contiguous I/Os to try to spot when the I/O is in one 60 of these sequential modes. But there are use-cases for wanting to 61 promote sequential blocks to the cache (e.g. fast application startup). 62 If sequential threshold is set to 0 the sequential I/O detection is 63 disabled and sequential I/O will no longer implicitly bypass the cache. 64 Setting the random threshold to 0 does _not_ disable the random I/O 65 stream detection. 66 67 Internally the mq policy determines a promotion threshold. If the hit 68 count of a block not in the cache goes above this threshold it gets 69 promoted to the cache. The read, write and discard promote adjustment 70 tunables allow you to tweak the promotion threshold by adding a small 71 value based on the io type. They default to 4, 8 and 1 respectively. 72 If you're trying to quickly warm a new cache device you may wish to 73 reduce these to encourage promotion. Remember to switch them back to 74 their defaults after the cache fills though. 75 76 Stochastic multiqueue (smq) 77 --------------------------- 78 79 This policy is the default. 80 81 The stochastic multi-queue (smq) policy addresses some of the problems 82 with the multiqueue (mq) policy. 83 84 The smq policy (vs mq) offers the promise of less memory utilization, 85 improved performance and increased adaptability in the face of changing 86 workloads. SMQ also does not have any cumbersome tuning knobs. 87 88 Users may switch from "mq" to "smq" simply by appropriately reloading a 89 DM table that is using the cache target. Doing so will cause all of the 90 mq policy's hints to be dropped. Also, performance of the cache may 91 degrade slightly until smq recalculates the origin device's hotspots 92 that should be cached. 93 94 Memory usage: 95 The mq policy uses a lot of memory; 88 bytes per cache block on a 64 96 bit machine. 97 98 SMQ uses 28bit indexes to implement it's data structures rather than 99 pointers. It avoids storing an explicit hit count for each block. It 100 has a 'hotspot' queue rather than a pre cache which uses a quarter of 101 the entries (each hotspot block covers a larger area than a single 102 cache block). 103 104 All these mean smq uses ~25bytes per cache block. Still a lot of 105 memory, but a substantial improvement nontheless. 106 107 Level balancing: 108 MQ places entries in different levels of the multiqueue structures 109 based on their hit count (~ln(hit count)). This means the bottom 110 levels generally have the most entries, and the top ones have very 111 few. Having unbalanced levels like this reduces the efficacy of the 112 multiqueue. 113 114 SMQ does not maintain a hit count, instead it swaps hit entries with 115 the least recently used entry from the level above. The over all 116 ordering being a side effect of this stochastic process. With this 117 scheme we can decide how many entries occupy each multiqueue level, 118 resulting in better promotion/demotion decisions. 119 120 Adaptability: 121 The MQ policy maintains a hit count for each cache block. For a 122 different block to get promoted to the cache it's hit count has to 123 exceed the lowest currently in the cache. This means it can take a 124 long time for the cache to adapt between varying IO patterns. 125 Periodically degrading the hit counts could help with this, but I 126 haven't found a nice general solution. 127 128 SMQ doesn't maintain hit counts, so a lot of this problem just goes 129 away. In addition it tracks performance of the hotspot queue, which 130 is used to decide which blocks to promote. If the hotspot queue is 131 performing badly then it starts moving entries more quickly between 132 levels. This lets it adapt to new IO patterns very quickly. 133 134 Performance: 135 Testing SMQ shows substantially better performance than MQ. 136 137 cleaner 138 ------- 139 140 The cleaner writes back all dirty blocks in a cache to decommission it. 141 142 Examples 143 ======== 144 145 The syntax for a table is: 146 cache <metadata dev> <cache dev> <origin dev> <block size> 147 <#feature_args> [<feature arg>]* 148 <policy> <#policy_args> [<policy arg>]* 149 150 The syntax to send a message using the dmsetup command is: 151 dmsetup message <mapped device> 0 sequential_threshold 1024 152 dmsetup message <mapped device> 0 random_threshold 8 153 154 Using dmsetup: 155 dmsetup create blah --table "0 268435456 cache /dev/sdb /dev/sdc \ 156 /dev/sdd 512 0 mq 4 sequential_threshold 1024 random_threshold 8" 157 creates a 128GB large mapped device named 'blah' with the 158 sequential threshold set to 1024 and the random_threshold set to 8.