14.5 Cluster File System Summary
We examined a number of cluster file systems and their architectures in this chapter. Although there is a lot of promise in PVFS, OpenGFS, Lustre, and the plethora of other open-source cluster file systems, the current developmental state of these file systems may not be appropriate for production environments. There are many more parallel file systems available from the open-source and research communities, and more are becoming available every day.
An alternative to the experimental nature of some of the parallel open-source file systems is good old NFS. We are mostly familiar with NFS and its characteristics, and should have a fairly good instinct about whether it is up to the challenge presented by our cluster. Like the other open-source cluster file systems, the price is right, but NFS has the advantage of being thoroughly wrung out by a large number of users who are familiar with it.
The commercially available file systems, like the Red Hat GFS and the PolyServe Matrix file system, are potentially viable alternatives to the open-source file systems for production environments that require scaling, resiliency, and availability of professional support. The support and feature set from the commercial cluster file systems makes them attractive for database, Web-serving, and high-availability cluster file system applications. It is unclear whether these file systems will scale into the many hundreds or thousands of clients.
Which choice you make for providing data to your cluster will depend on your ability to deal with software development issues, scaling, implementation details, and the availability of support. I recommend a thorough investigation before selecting a cluster file system for your environment. The good thing is that we have a wide range of choices.