[Gluster-devel] Client side afr versus server side, doing a self-heal

Anand Avati avati at zresearch.com
Thu May 1 15:24:21 UTC 2008


2008/5/1 Brandon Lamb <brandonlamb at gmail.com>:

> Would I be correct in this thinking?
>
> 2 servers as data nodes, multiple clients doing afr client side. This
> is a new setup where server1 has 100 gigs of data and server2 has
> empty directory.
>
> So by doing afr client side, does the client have to read and transfer
> data FROM server1, and then copy it to server2? Would doing server
> side afr be faster for this process? Or does server1 magically know to
> send the file directly to server2? Im assuming no, because from my
> understanding server1 and server2 dont even know about each other as
> its the client doing the afr?
>
> I guess there are multiple ways to get your initial setup, set up as
> well? What would you recommend in a situation like this? Setup server
> side afr at first to get the data to server2 then take down the
> cluster and set it up as client side afr once all the data is copied?
> Or just go with the data having to be copied over the network twice
> (if this is how it works)
>

One easy way is to have one of the servers have a mount point (client) and
convert the network traffic to loopback traffic (i.e., if you absolutely do
not want to change the client spec file temporarily)

avati



More information about the Gluster-devel mailing list