On 03/02/2012, Tom Rosmond <rosm...@reachone.com> wrote: > Recently the organization I work for bought a modest sized Linux cluster > for running large atmospheric data assimilation systems. In my > experience a glaring problem with systems of this kind is poor IO > performance. Typically they have 2 types of network: 1) A high speed, > low latency, e.g. Infiniband, network dedicated to MPI communications, > and, 2) A lower speed network, e.g 1Gb or 10Gb ethernet, for IO. On > clusters this second network is usually the basis for a global parallel > file system (GPFS), through which nearly all IO traffic must pass.
Not necessarily. You can use the Infiniband network for storage very easily, and this is commonly done. Indeed, you can increase your bandwidth for storage by using the two ports of the Infiniband adapters, allocating even and odd compute nodes to different storage subnets. All you need do is have an Infiniband card (or cards) in your storage server, or to put in dedicated storage routers (such as Panasas do). GPFS is not the only kid on the block. There are other filesystems out there - such as Panasas and Lustre. And indeed what is wrong with having a properly balanced RAID array and an NFS server with an Infiniband interface (or interfaces). The 'glaring problem' is not inherent in clusters.
