[Beowulf] Doing i/o at a small cluster

[Andrew Holway]

> So some sort of distributed file system seems the best option, and a
> lot cheaper and a lot faster than a dedicated fileserver
> that will not be able to keep up.

a) ZFS doesn't use hardware raid. at all. ever.

b) 500MB/s (actually 1GB/s) of I/O will chew up a quite large amount
of resource. Just think of all that nasty nasty TCP that your going to
have to do. Your essentially doubling the I/O requirements of each
node. Also, most distributed filesystems have a quite heavy CPU
overhead. Recalculating hashes and the such like is dirty work.

c) ZFS would use one or two SSD's as a read cache. Having an SSD only
volume is a quite rustic way of doing things nowadays.

d) I think the only tool that does what your describing, gluster,
really sucks because it is still FUSE.

e) ZFS is not yet on linux. Nexenta is based on what used to be OpenSolaris.

f) It does sound like a good fit for hadoop actually.

ta for now.

>
>
>
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf