]> git.apps.os.sepia.ceph.com Git - ceph.git/commitdiff
doc/rados: Better block.db size recommendations for bluestore 32226/head
authorNeha <nojha@redhat.com>
Thu, 12 Dec 2019 22:04:57 +0000 (22:04 +0000)
committerNeha <nojha@redhat.com>
Fri, 13 Dec 2019 17:24:58 +0000 (17:24 +0000)
Signed-off-by: Neha Ojha <nojha@redhat.com>
doc/rados/configuration/bluestore-config-ref.rst

index d728775d46adfe9fe9171b502f920d3182497ac7..bcc8476ab8d9cca738e98664e512681407df10a2 100644 (file)
@@ -127,9 +127,11 @@ When using a :ref:`mixed spinning and solid drive setup
 ``block.db`` logical volume for Bluestore. Generally, ``block.db`` should have
 *as large as possible* logical volumes.
 
-It is recommended that the ``block.db`` size isn't smaller than 4% of
-``block``. For example, if the ``block`` size is 1TB, then ``block.db``
-shouldn't be less than 40GB.
+The general recommendation is to have ``block.db`` size in between 1% to 4%
+of ``block`` size. For RGW workloads, it is recommended that the ``block.db``
+size isn't smaller than 4% of ``block``, because RGW heavily uses it to store its
+metadata. For example, if the ``block`` size is 1TB, then ``block.db`` shouldn't
+be less than 40GB. For RBD workloads, 1% to 2% of ``block`` size is usually enough.
 
 If *not* using a mix of fast and slow devices, it isn't required to create
 separate logical volumes for ``block.db`` (or ``block.wal``). Bluestore will