[Gluster-users] Looking for configuration suggestions

Matt Lawrence matt.lawrence at tamu.edu
Mon Oct 27 14:37:30 UTC 2008


Vikas Gorur wrote:
> Good to hear. It'll help us to recommend a configuration to you
> if we know more about how you're going to use the storage cluster.
>   

Mainly for storing meteorlogical data, including lots of hurricane 
information.

I expect to have 50TB under glusterfs in the next couple of weeks and 
probably 100TB by the end of the year.

> It makes sense to split up your storage into multiple bricks, atleast one
> per disk. This way access can proceed concurrently on many disks.
>   
Not really a viable option since each unit after being turned into a 
RAID 5 is about 12TB.  Trying to run them as raw disks would probably be 
a bad idea.  However, I can split up the 12TB into multiple LUNs to be 
presented to the server.  The storage will still be spread across  the 
same disks, so there won't be any  performance gain from being able to 
access disks independently.  However,  I'm wondering if there is an 
advantage to keeping the brick size down.

> Configuring GlusterFS for optimum performance would have to be done 
> knowing
> the kind of load you expect. What sort of applications will be running?
> What are their data access patterns? (mostly reads? mostly writes? small files?
> huge files? etc.)
>   

WRF is probably the biggest user with some other wave prediction models 
running second.  So, I expect there are a number of large files.  Given 
that these are connected via Gb ethernet, I expect performance to be 
more network limited than anything else.

I expect replication to be implemented sometime in the future, probably 
to another site.

Again, thanks for the help!

-- Matt




More information about the Gluster-users mailing list