a) the master node sends jobs to the slave nodes, the nodes reply the results to the master node, and the master writes down the output files
b) the master node sends jobs to the slave nodes, and each nodes write its own piece of output file.
We are building a new cluster with a SAN FC disks, and we are thinking about configuring a file system clustered in the SAN partition.
But if the model runs with configuration a), we think we can not expect a significant growth in the performance compared with NFS, since the limit should be related to the ethernet bandwidth... at least, this is what I think I understood.
Is anyone keen on this ready to share some info?
