Lexington Luthor writes:
> David Masover wrote:
> > Great -- no load balancing or local caching at all, then?
> >
>
> Its actually pretty fast - almost raw disk speed for small clusters, but
> not much slower for larger ones - try it out.
Lustre and GFS target quite different environments. GFS is mainly for
small-to-medium clusters with up to hundred nodes, for situations when
one needs fail-over and things like that. Lustre is for the large
clusters with tens thousand nodes, hundreds object stores, with very
fa{,s}t transports both through the network (elan) and to the storage
(FC). This requires completely different architecture of IO and
meta-data management.
>
> Its not as fast as reiser5 may end up, but its here and working right
> now. It is load balanced, and it does local caching. Use it before you
> start criticizing it.
>
> It has plenty of its own problems (mainly relating to the DLM), but none
> of the things you mentioned are a problem.
>
> For the time being, it beats every other non-commercial distributed file
> system hands down.
>
> I look forward to reiser5, but GFS is just fine for now.
>
> LL
>
Nikita.
>
