[Gluster-users] GlusterFS performance

[David Seira]

Hi list,

I'm trying to deploy a GlusterFS cluster to use as a base filesystem
storage for VMWare ESXi. I want to test the glusterFS performance to store
the virtual machines.

I've created two VMs with only two replicas, both VM have 16GB of RAM and 8
CPU at 2 GHz. I've configured the glusterFS volume with the default
parameters. The replication works perfectly; every file is replicated
automatically. For the gluster volume I've used an SSD disk (Kingston
HyperX SSD 3K 120GB with reads at 555MB/s, writes at 510MB/s).

To test the glusterFS performance, I've configured the GlusterFS cluster as
a storage in another ESXi through NFS. Then I've created a ubuntu virtual
machine in this ESXi.

For the test, I've used the fio and dd tool. The maximum IOPS I've achieved
using fio for random-write is 2000 and a 70MB/s of transfer rate. To
compare the results I've installed a VM directly into the SSD (without
GlusterFS interaction) and the results are 29K IOPS and 120MB/s. In
sequencial access the result are similar.

I don't know why the performance is so reduced.

Are there any special option to configure the replicate mode to increase
this behaviour (big files with a lot of writes)?

I've also tested the volume with this options:

nfs.trusted-write: On
performance.write-behind-window-size: 1GB
performance.io-thread-count: 64
performance.cache-size: 1GB
performance.flush-behind: on
performance.write-behind: on
performance.low-prio-threads: 64
performance.least-prio-threads: 64
performance.high-prio-threads: 64
performance.normal-prio-threads: 64
performance.nfs.write-behind: on

But with similar results.

Any of you have tested GlusterFS as a storage for the ESXi?

Best regards,
David
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130927/78b010af/attachment.html>