In a very few situations raw gives a
little more performance but in most of cases qcow2 is good enough,
we're using qcow2 by default: it allows snap-shooting and thin
provisioning.
If you have redundant PSU's you can use writeback or writethrough (safest). No need for directsync with xfs in vm's. Remember to switch disk queue scheduler to none or at least deadline. Some gluster parameters we use in production for vm's: gluster vol set vmstore features.shard-block-size 256MB; gluster vol set vmstore features.shard enable; gluster vol set vmstore performance.read-ahead off; gluster vol set vmstore performance.quick-read off; gluster vol set vmstore performance.io-cache off; gluster vol set vmstore performance.stat-prefetch off; gluster vol set vmstore cluster.data-self-heal-algorithm full; gluster vol set vmstore client.event-threads 2; gluster vol set vmstore server.event-threads 2; gluster vol set vmstore performance.client-io-threads on; gluster vol set vmstore network.remote-dio enable; gluster vol set vmstore performance.force-readdirp false; gluster vol set vmstore dht.force-readdirp false; El 11/09/19 a les 15:04, Cristian Del Carlo ha escrit:
|
_______________________________________________ Gluster-users mailing list Gluster-users@xxxxxxxxxxx https://lists.gluster.org/mailman/listinfo/gluster-users