[ceph-users] Sizing for bluestore db and wal

[Felix Stolte]

Hi folks,

i upgraded our ceph cluster from jewel to luminous and want to migrate
from filestore to bluestore. Currently we use one SSD as journal for
thre 8TB Sata Drives with a journal partition size of 40GB. If my
understanding of the bluestore documentation is correct, i can use a wal
partition for the writeahead log (to decrease write latency, similar to
filestore) and a db partition for metadata (decreasing write AND read
latency/throughput). Now I have two questions:

a) Do I really need an WAL partition if both wal and db are on the same SSD?

b) If so, what would the ratio look like? 99% db, 1% wal?


Best regards, Felix

[Janne Johansson]

Post by Felix Stolte
Hi folks,
i upgraded our ceph cluster from jewel to luminous and want to migrate
from filestore to bluestore. Currently we use one SSD as journal for
thre 8TB Sata Drives with a journal partition size of 40GB. If my
understanding of the bluestore documentation is correct, i can use a wal
partition for the writeahead log (to decrease write latency, similar to
filestore) and a db partition for metadata (decreasing write AND read
a) Do I really need an WAL partition if both wal and db are on the same SSD?

I think the answer is no here, if you point the DB to an SSD, bluestore will
use it for WAL also.

Post by Felix Stolte
b) If so, what would the ratio look like? 99% db, 1% wal?

..which means just let ceph handle this itself.