For OSDs like ZFS to perform optimally it's import that they be allowed
to manage their own cache. This maximizes the likelyhood that the ARC
will prefetch and cache the right buffers. In the existing ZFS OSD
code a cached LU object pins buffers in the ARC preventing them from
being dropped. As the LU cache grows it can consume the entire ARC
preventing buffers for other objects, such as the OIs, from being
cached and severely impacting the performance for FID lookups.
By default this patch will only limit the LU cache for ZFS OSDs.
NOTES:
* Setting LU_CACHE_NR_ZFS_LIMIT to 0 results in an LBUG on the MDS.
This may be because an object is being used without a reference.
Setting a minimum value of 256 was arbitrary, ideally we would
set this value to 0.
* In order to be able to quickly determine the number of objects in
the hash table the CFS_HASH_COUNTER flag is added. This adds an
atomic_inc/dec to the hash insert/remove paths but is not expected
to have any measurable impact of performance.
This patch is back-ported from the following one:
Lustre-commit:
0123baecc4e2050447f8c4f48f5b33a6d3c524a8
Lustre-change: http://review.whamcloud.com/10237
Test-Parameters: alwaysuploadlogs \
mdtfilesystemtype=zfs mdsfilesystemtype=zfs ostfilesystemtype=zfs \
testgroup=review-zfs
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Change-Id: If573f02175a02abeeb100ed74f3109e629d4c392
Reviewed-on: http://review.whamcloud.com/12001
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Reviewed-by: Isaac Huang <he.huang@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>