[Gluster-users] replication is 5x slower for dht?

Sacerdoti, Federico Federico.Sacerdoti at DEShawResearch.com
Thu May 21 21:42:31 UTC 2009


I updated to gluster/2.0.1 which fixed my issues with getting
cluster/replicate to work, thanks. It now does, and I can write 32GB
from 1-80 writers to 10 gluster servers with no errors.

However the replicated performance is 5x lower than the non-replicated
DHT/distribute run. One factor is the replicated runs have half the
number of subvolumes. However the overhead of replication still seems

Question: I see the features/locks translator is required for replicate.
Why is this needed? Could it be contributing to the performance


More information about the Gluster-users mailing list