[Gluster-users] tuning

Papp Tamas tompos at martos.bme.hu
Thu Jul 7 14:01:49 UTC 2011

On 2011-07-07 15:42, Joe Landman wrote:
>> The same on the cluster volume is ~50-60 MB/s.
> This is what you expect over GE, more towards the lower end of the 
> range (assuming a non-optimized driver and network stack).

Why? It's GE, 60MB/s is about ~500Mbit. I expect to be at least 
~900Mbit. Am I dreaming?:)

>> Network layer is GE, nodes are connected with two NICs in bonding.
> Bonding will not help single threaded reads/writes.  It could help 
> multiple simultaneous reads and writes.  But a single process doing 
> reads/writes will go over a single link (in general).

You are right. I just wrote it to be sure about writing every details.

>> Or does the Storage Platform run on an optimized kernel or something
>> like that?
> I am absolutely desparated. Is it Ubuntu? Would be better with Fedora?
> I don't understand this.  What are your needs, and what are your 
> expectations based upon those needs?  Do you have the right equipment 
> to meet the needs, or do you need to buy more/better equipment to meet 
> the needs?

800-1000Mbit/s would be enough to me.
Actually I had never tested, that would be the optimal for our purposes. 
We do the best from the tools we have, and if it's not enough, the boss 
starting to think over it...
I know, this is not the best way, but we have this:)

What can be the bottleneck for our system, GE?
Would 10GE for server help on this much more (single machines to cluster 

Thank you,


ps.: What do you mean on optimizing network stack, jumbo frame?

More information about the Gluster-users mailing list