TEXT   73

cache-policies.txt

Guest on 25th July 2021 05:24:15 PM

  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
  29. ----------
  30.  
  31. This policy is the default.
  32.  
  33. The multiqueue policy has two sets of 16 queues: one set for entries
  34. waiting for the cache and another one for those in the cache.
  35. Cache entries in the queues are aged based on logical time. Entry into
  36. the cache is based on variable thresholds and queue selection is based
  37. on hit count on entry. The policy aims to take different cache miss
  38. costs into account and to adjust to varying load patterns automatically.
  39.  
  40. Message and constructor argument pairs are:
  41.         'sequential_threshold <#nr_sequential_ios>' and
  42.         'random_threshold <#nr_random_ios>'.
  43.  
  44. The sequential threshold indicates the number of contiguous I/Os
  45. required before a stream is treated as sequential.  The random threshold
  46. is the number of intervening non-contiguous I/Os that must be seen
  47. before the stream is treated as random again.
  48.  
  49. The sequential and random thresholds default to 512 and 4 respectively.
  50.  
  51. Large, sequential ios are probably better left on the origin device
  52. since spindles tend to have good bandwidth. The io_tracker counts
  53. contiguous I/Os to try to spot when the io is in one of these sequential
  54. modes.
  55.  
  56. cleaner
  57. -------
  58.  
  59. The cleaner writes back all dirty blocks in a cache to decommission it.
  60.  
  61. Examples
  62. ========
  63.  
  64. The syntax for a table is:
  65.         cache <metadata dev> <cache dev> <origin dev> <block size>
  66.         <#feature_args> [<feature arg>]*
  67.         <policy> <#policy_args> [<policy arg>]*
  68.  
  69. The syntax to send a message using the dmsetup command is:
  70.         dmsetup message <mapped device> 0 sequential_threshold 1024
  71.         dmsetup message <mapped device> 0 random_threshold 8
  72.  
  73. Using dmsetup:
  74.         dmsetup create blah --table "0 268435456 cache /dev/sdb /dev/sdc \
  75.             /dev/sdd 512 0 mq 4 sequential_threshold 1024 random_threshold 8"
  76.         creates a 128GB large mapped device named 'blah' with the
  77.         sequential threshold set to 1024 and the random_threshold set to 8.

Raw Paste


Login or Register to edit or fork this paste. It's free.