Craig Cory wrote: > Perhaps a better question is what does GlusterFS give me over the > Lustre FS that Sun acquired a little while back? I'm not well versed > in either, but my understanding is the "concepts" are similar. > > What platform/architecture is GlusterFS geared to? Lustre is > currently Linux/x86 based. > > Thanks. >
And thanks for asking this question. There are many differences when we compare GlusterFS to Lustre. The most prominent one being that GlusterFS is a user-space application. The advantage that comes with that is the ability to work on any POSIX compatible operating system. It is also architecture-agnostic. From a design point of view, GlusterFS provides the benefit of not having a meta-data server, unlike Lustre. GlusterFS aim has been to completely avoid the potential of single points of failures that come with meta-data servers. On the storage side, I believe, and please correct me if I am wrong, Lustre requires a specific file system layout on the storage nodes or OSTs as they are called. GlusterFS has no such requirement. It works out of the box over any on-disk file system that supports extended attributes. As compared to an in-kernel cluster file system like Lustre, it is far far simpler to setup GlusterFS, especially considering the overheads for users when dealing with a Linux kernel that does not include Lustre in the mainline. You're welcome to hop over to the gluster-users list for more information. See http://gluster.org/mailing-list.php or the #gluster IRC channel on irc.gnu.org. Regards Shehjar > > Shehjar Tikoo wrote: >> Jonathan Adams wrote: >>> What advantages does this give us over ZFS systems (which can do >>> NFS/CIFS native) >> The biggest advantage, compared to NFS/CIFS, is the ability to >> scale out the storage deployment beyond a single or a few >> servers/bricks/racks. >> >> GlusterFS enables this because the design is such that performance >> and storage of multiple nodes can be aggregated into a single >> name-space while ensuring that management overhead/complexity >> remains low. >> >> It provides safety against node failures through in-built >> replication ,striping and name-space distribution. Spreading out >> the data over multiple nodes brings with it the benefits of >> reducing hot-spots in workloads and access patterns. >> >> However, when dealing with networked storage, there is always the >> possibility of network failures and partitions. GlusterFS provides >> multi-pathing and high-availability functionality that ensures >> clients can work without disruption. The replication functionality >> is also able to handle such partitions and split-brains by >> performing self-heal automatically. >> >> New users dont have to format the disks either. Since GlusterFS >> works in user-space, it can use any POSIX compatible file system in >> one server and create a storage cluster together with completely >> different file systems on other nodes. >> >> We even have GlusterFS being used on production systems over ZFS on >> Solaris. >> >> Regards Shehjar _______________________________________________ >> nfs-discuss mailing list nfs-discuss at opensolaris.org >> > >
