On 08/12/13 01:34, Jung Young Seok wrote: > Dear Gluster Users, > > We're currently replacing storage solution to GlusterFS > and trying to find bottleneck so we can utilize GlusterFS as expected. > > One thing I'd like to ask is about glusterfs client's broadcasting. > > We have 2 gluserfs servers and each server has 2 bricks. > Each brick replicates to other side. (So we use logically 2 bricks) > > When I mount it as glusterfs and copy a file on the mounted folder, > the client sends the file to both server (to replicate each other) > > I want the replication to be done in glusterfs server not in client side. > > However, if I mount glusterfs node as nfs, replication is done in server > side. NFS is how that is accomplished. Though you will want to figure out some HA method for the NFS IP address. > > I'd like to avoid using client side 10g network and I'd like to use 10g > on Storage server side if it should be. > > Is there any way let the replication job done in glusterfs server side > not in client side? You are not really clear on this point. I am guessing you have the storage servers connected to two separate networks, via two separate 10G network interfaces, correct? If that is correct, then the most popular way is to make sure all storage servers are known to each other via host names, and not IP addresses. (peer probe <hostname>) Then you would have enteries in the host table (/etc/hosts) on each storage server, that list all of the storage servers using the IP address of the interfaces you want the replication traffic to go across. > (We're using CentOS5.8 for client[WAS], CentOS6.3 for GlusterFS Server) > > If I broke the mailing rule, please understand and explain how I can use > it. (I Couldn't use the IRC caht) This is what the mailing list is here for. (IRC is just another option.) -- Mr. Flibble King of the Potato People
