Skip to content

L
linux
Switch branch/tag
  1. 26 Jun, 2015 2 commits
  3. 17 Jun, 2015 6 commits
  5. 11 Jun, 2015 12 commits
    • Joe Thornber's avatar
      dm thin metadata: fix a race when entering fail mode · b1f11aff
      Joe Thornber authored 
      In dm_thin_find_block() the ->fail_io flag was checked outside the
metadata device's root_lock, causing dm_thin_find_block() to race with
the setting of this flag.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      b1f11aff
    • Mike Snitzer's avatar
      dm thin: fail messages with EOPNOTSUPP when pool cannot handle messages · fd467696
      Mike Snitzer authored 
      Use EOPNOTSUPP, rather than EINVAL, error code when user attempts to
send the pool a message.  Otherwise usespace is led to believe the
message failed due to invalid argument.
Reported-by: default avatarZdenek Kabelac <zkabelac@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      fd467696
    • Joe Thornber's avatar
      dm thin: range discard support · 34fbcf62
      Joe Thornber authored 
      Previously REQ_DISCARD bios have been split into block sized chunks
before submission to the thin target.  There are a couple of issues with
this:

 - If the block size is small, a large discard request can
   get broken up into a great many bios which is both slow and causes
   a lot of memory pressure.

 - The thin pool block size and the discard granularity for the
   underlying data device need to be compatible if we want to passdown
   the discard.

This patch relaxes the block size granularity for thin devices.  It
makes use of the recent range locking added to the bio_prison to
quiesce a whole range of thin blocks before unmapping them.  Once a
thin range has been unmapped the discard can then be passed down to
the data device for those sub ranges where the data blocks are no
longer used (ie. they weren't shared in the first place).

This patch also doesn't make any apologies about open-coding portions
of block core as a means to supporting async discard completions in the
near-term -- if/when late bio splitting lands it'll all get cleaned up.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      34fbcf62
    • Joe Thornber's avatar
      dm thin metadata: add dm_thin_remove_range() · 6550f075
      Joe Thornber authored 
      Removes a range of blocks from the btree.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      6550f075
    • Joe Thornber's avatar
      dm thin metadata: add dm_thin_find_mapped_range() · a5d895a9
      Joe Thornber authored 
      Retrieve the next run of contiguously mapped blocks.  Useful for working
out where to break up IO.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      a5d895a9
    • Joe Thornber's avatar
      dm btree: add dm_btree_remove_leaves() · 4ec331c3
      Joe Thornber authored 
      Removes a range of leaf values from the tree.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      4ec331c3
    • Pekka Enberg's avatar
      dm stats: Use kvfree() in dm_kvfree() · 0f24b79b
      Pekka Enberg authored 
      Use kvfree() instead of open-coding it.
Signed-off-by: default avatarPekka Enberg <penberg@kernel.org>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      0f24b79b
    • Joe Thornber's avatar
      dm cache: age and write back cache entries even without active IO · fba10109
      Joe Thornber authored 
      The policy tick() method is normally called from interrupt context.
Both the mq and smq policies do some bottom half work for the tick
method in their map functions.  However if no IO is going through the
cache, then that bottom half work doesn't occur.  With these policies
this means recently hit entries do not age and do not get written
back as early as we'd like.

Fix this by introducing a new 'can_block' parameter to the tick()
method.  When this is set the bottom half work occurs immediately.
'can_block' is set when the tick method is called every second by the
core target (not in interrupt context).
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      fba10109
    • Mike Snitzer's avatar
      dm cache: prefix all DMERR and DMINFO messages with cache device name · b61d9509
      Mike Snitzer authored 
      Having the DM device name associated with the ERR or INFO message is
very helpful.
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      b61d9509
    • Joe Thornber's avatar
      dm cache: add fail io mode and needs_check flag · 028ae9f7
      Joe Thornber authored 
      If a cache metadata operation fails (e.g. transaction commit) the
cache's metadata device will abort the current transaction, set a new
needs_check flag, and the cache will transition to "read-only" mode.  If
aborting the transaction or setting the needs_check flag fails the cache
will transition to "fail-io" mode.

Once needs_check is set the cache device will not be allowed to
activate.  Activation requires write access to metadata.  Future work is
needed to add proper support for running the cache in read-only mode.

Once in fail-io mode the cache will report a status of "Fail".

Also, add commit() wrapper that will disallow commits if in read_only or
fail mode.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      028ae9f7
    • Joe Thornber's avatar
      dm cache: wake the worker thread every time we free a migration object · 88bf5184
      Joe Thornber authored 
      When the cache is idle, writeback work was only being issued every
second.  With this change outstanding writebacks are streamed
constantly.  This offers a writeback performance improvement.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      88bf5184
    • Joe Thornber's avatar
      dm cache: add stochastic-multi-queue (smq) policy · 66a63635
      Joe Thornber authored 
      The stochastic-multi-queue (smq) policy addresses some of the problems
with the current multiqueue (mq) policy.

Memory usage
------------

The mq policy uses a lot of memory; 88 bytes per cache block on a 64
bit machine.

SMQ uses 28bit indexes to implement it's data structures rather than
pointers.  It avoids storing an explicit hit count for each block.  It
has a 'hotspot' queue rather than a pre cache which uses a quarter of
the entries (each hotspot block covers a larger area than a single
cache block).

All these mean smq uses ~25bytes per cache block.  Still a lot of
memory, but a substantial improvement nontheless.

Level balancing
---------------

MQ places entries in different levels of the multiqueue structures
based on their hit count (~ln(hit count)).  This means the bottom
levels generally have the most entries, and the top ones have very
few.  Having unbalanced levels like this reduces the efficacy of the
multiqueue.

SMQ does not maintain a hit count, instead it swaps hit entries with
the least recently used entry from the level above.  The over all
ordering being a side effect of this stochastic process.  With this
scheme we can decide how many entries occupy each multiqueue level,
resulting in better promotion/demotion decisions.

Adaptability
------------

The MQ policy maintains a hit count for each cache block.  For a
different block to get promoted to the cache it's hit count has to
exceed the lowest currently in the cache.  This means it can take a
long time for the cache to adapt between varying IO patterns.
Periodically degrading the hit counts could help with this, but I
haven't found a nice general solution.

SMQ doesn't maintain hit counts, so a lot of this problem just goes
away.  In addition it tracks performance of the hotspot queue, which
is used to decide which blocks to promote.  If the hotspot queue is
performing badly then it starts moving entries more quickly between
levels.  This lets it adapt to new IO patterns very quickly.

Performance
-----------

In my tests SMQ shows substantially better performance than MQ.  Once
this matures a bit more I'm sure it'll become the default policy.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
      66a63635
  7. 29 May, 2015 20 commits
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7