[Gluster-users] performance issues of gluster
hlv at redhat.com
Fri Nov 2 12:56:48 UTC 2012
I'v got a Gluster testing, and found some performance issuse pertain to Gluster
I built a ten-node Gluster enviroment, (3 X HP 380G5, 4 IBM 3650, 2 X HP C7000. 1 X DELL PE2950, all the network switch & NICs are all gigabit).
And set up three volumes:
a ten-brick distribute volume (ten-brick means this volumes resides on ten nodes, others are same),
a five-brick distribute volume,
a ten-brick distribute replica volume
And I copied a folder ( including about 100 files, the size of everyfile is 40MB~80MB) to these three volumes one by one, and found the average speeds are:
to ten-brick distribute volume : 25MB/s
to five-brick distribute volume: 50MB/s
to ten-brick distribute replica volume: 10MB/s
Who have any idea about:
Why the performance is so bad?
Why the performance of ten-brick distribute volume is worse than five-brick's(technically ten-brick volume should be better than five-brick volume)?
How to tune it?
More information about the Gluster-users