Re: Distributed filesystems in Debian
Brett Viren wrote:
> About a month ago the issue of distributed filesystems in Debian was
> raised here. Since then, has anyone had any experiences, good or bad?
>
> Personally, I am looking for a good way to serve 16 disks (8TB), split
> into two nodes to a cluster of 18 nodes.
>
> My requirements (or maybe better "desires") are:
>
> - present a single, monolithic as possible large filesystem
>
> - enough redundancy so one dead disk doesn't kill the whole
> filesystem's files
>
> - RAID0 like parallelism to avoid bottlenecks
>
> - Free Software (Debian packages best), simple install and
> maintenance.
>
> - Good match to my cluster size (10s of nodes), additional hardware
> not required.
>
> - Makes my morning coffee for me.
>
> Currently I'm leaning towards using Lustre, but I worry it may not be
> a good fit to the small size of my cluster. I'd also enjoy hearing
> about the applicability of PVFS2 and Red Hat's GFS.
>
> -Brett.
>
>
>
For the current versions of Lustre (1.4.9 or 1.5.97, in Debian Unstable),
use Lustre on RAID1 or RAID5 to provide redundancy. Lustre-RAID
(or LAID as its called in the documentation), is one the Lustre roadmap
for this year, but not there yet.
For performance bottlenecks, Lustre wins, but its poor for small files;
its optimised for HPC at the moment, but there is some work by
Cisco folks mentioned on the lustre-discuss mailing lists; they are
running lustre from 512 MB / 1 GB flash devices!
As for making the coffee, patches are welcome :-)
Regards
Alastair McKinstry, debian lustre maintainer.
Reply to: