[Gluster-users] Recommended volume settings for 10G VMWare Datastore

[Ron Kelley]

Greetings all,

I am setting up a new pair of servers in the data center to be used
exclusively as Gluster storage servers for VMWare.  Each server is
identical with 16G RAM, dual 10G NICs, 12x 600G 15K-RPM SAS drives, and an
LSI RAID controller. Each will be running CentOS 6.5 with the latest
patches and the 3.16 ML kernel as well as gluster 3.5.2 from the standard
RPM packages.

I have done a ton of research on gluster and have been testing it heavily
in my development environment.  I created a new replicated volume using a
single 3TB brick per server, and the performance has been very good so far.
I spun up a few Win7 test VMs running IOMeter and can easily hit +100MB/sec
read/writes on each client.  I have also run through some failover
scenarios using ucarp (power off conditions, etc), and Gluster has worked
very well!

Unfortunately, my lab only has a couple of Hypervisor hosts, and I have not
been able to load up a bunch of VMs to do a full-scale test.  Thus, I am
wondering if I should modify any volume settings to ensure Gluster
continues to perform this well once we go to production.

>From previous emails on this mailer list, I see some people have adjusted
things like  network.ping-timeout, performance.cache-size,
performance.cache-[min|max]-file-size,
performance,write-behind-window-size, etc.

In addition, I found this URL which points to some undocumented performance
settings:
http://www.gluster.org/community/documentation/index.php/Documenting_the_undocumented#Performance


Can someone please share their thoughts/best-practices when using Gluster
to host datastore files in a 10G network environment?

Thanks in advance,

-Ron
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20140909/941ce939/attachment.html>