[Gluster-users] Mixing transport-types?

Nikola Garafolic nikola.garafolic at srce.hr
Thu Mar 25 12:32:38 UTC 2010

I have 4 gluster storage nodes, they have both IB and ETH. Since this 
gluster storage should be used for cluster, problem is that cluster 
nodes have IB and cluster frontend has ETH. So in theory cluster nodes 
could mount gluster via IB, and frontend via ETH, right?

Running two or more gluster clients with different transport type (if 
gluster storage nodes are set to run via IB and ETH) will not cause any 


Tejas N. Bhise wrote:
> Nikola,
> Can you please provide more information on which of these are identified to be gluster servers ( exporting backend data volumes ) and which one of these are client nodes ( mounting volumes exported by gluster servers ). The gluster clients where the mounts are done are usually application servers.
> Also tell us more about whether you want replication or not for your data. You can take three servers and put  together in a distribute or distribute-replicate config and mount on the fourth server using all IB.
> Or you would provide failover for the mount and have two servers and two clients and still use distribute-replicate.
> If the client is a node outside of these 4 nodes and use TCP, then the glusterfs server has to export data using tcp protocol and not IB since the client is using TCP.
> The server can export using both tcp and ib - that is the example you pointed to.
> Since the glusterfs servers dont talk to each other, if you are planning on using all 4 machines as servers, their IB connection to each other is a waste, unless there is other data, outside of glusterfs, that you want to move around between them. Might as well use that to connect to some of the clients which are going to run important application servers ( just a suggestion ).
> Feel free to ask more questions.
> Regards,
> Tejas.

More information about the Gluster-users mailing list