Hi Francois, On Thu, Dec 24, 2009 at 5:22 AM, Francois Berenger <berenger at riken.jp>wrote: > Hello, > > I plan to do a small compiling farm. > What would be the best architecture and mode to use? > > Initially, I was planning to have one server (which is also > a client) and three other machines which are only clients. > > I am asking myself if having all machines acting both as > client and server wouldn't be better. > > What I want to avoid is that when a file is read from the server > the bandwidth is shared between clients to access this file. > I'd like to have clients possessing a file help the server to distribute > it to other nodes rather than having the server only serving files. > I don't know if this is the default behavior of GlusterFS. > you can obtain this behavior by using replicate. 3 servers can be run on all the nodes and clients running on each of these 3 nodes can have a replicate of 3 children - 3 protocol/client speaking to each of the server - and each of the client specifying the "read-subvolume" as local server in replicate. But since replicate replicates all the data, think of whether you can afford to maintain 3 replicas of data. > > The usage pattern would be like this I guess: all clients read almost > all files from the source tree, each client produce and write a few files. > > Thank you, > Francois. > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://gluster.org/cgi-bin/mailman/listinfo/gluster-users > regards, -- Raghavendra G
