>I am getting some problems understanding the actual >idea behind Cluster File System implementation. From >one website >http://www.stalker.com/notes/SFS.html#CFS >I got the idea that a cluster file system is a wrapper >file system. >... >On the other hand OpenGFS >http://opengfs.sourceforge.net/ >which is also a cluster file system has its own format >utility and disk layout. > >Now what is exactly the definition of a cluster file >system. Maybe you missed the word "usually," where that web page says that a cluster file system is usually a wrapper around some other filesystem. I can't comment on whether "usually" is a fair description, but at least it resolves your inconsistency. The definition of a cluster file system is quite simple: A file system appropriate for use in a cluster. A cluster is a bunch of computers working on the same problem (usually in order to solve it faster than a single computer could). The thing about a cluster that is incompatible with many file systems is that because the computers are working on the same problem, they probably need to read and write the same files, at the same time. Something like ext3 doesn't allow multiple computers to have the same filesystem mounted at the same time (unless read-only). A "shared" filesystem _can_ be mounted on multiple computers at the same time, but may not be a cluster file system because it doesn't perform acceptably when multiple computers are accessing (especially writing) to the same file at the same time. NFS, for example, is a shared filesystem type that usually isn't fast enough to satisfy a cluster. So sometimes, the cheapest way to make a cluster filesystem is to wrap something around a non-cluster filesystem, but sometimes you want to build a filesystem for clustering from the ground up. While we're talking filesystem class definitions, I'll throw in another that you often hear along with cluster and shared: distributed filesystem. While I've seen it used to mean "shared," what it really means is that the filesystem, or at least the entity through which you access it, is spread across multiple computers. For example, two computers might each own a disk and the two disks together form a filesystem. Or maybe the disks don't belong to any particular computer, but the locks that coordinate access to them are spread across multiple computers. It is in the nature of cluster computing for a cluster to use a distributed filesystem (i.e. the filesystem itself is clustered), but it isn't required. -- Bryan Henderson IBM Almaden Research Center San Jose CA Filesystems - To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
