Re: [myrinet] distributed file system

I doubt the complexity of NFS is exactly what you need.  Perhaps PVFS
is a better choice:
http://parlweb.parl.clemson.edu/pvfs

And here's a paper from the PVFS guys describing the filesystem.  In
the 'Performance' section, they discuss bandwitdh over myrinet:
http://parlweb.parl.clemson.edu/pvfs/el2000/extreme2000.html


On Mon, Oct 09, 2000 at 12:45:58PM +0200, Alan Scheinine wrote:
> What is the fastest way to access files on a remote disk using Myrinet?
> This is a very broad question and I will try to narrow it.  One can
> consider NFS or specialized file systems that keep a local copy.
> I tried a distributed file system under development, coda, but (for
> me) it was a headache, too many problems arose that required manual
> intervention.  While I would be interested in hearing about various
> file systems -- for example, I am interested in the status of Andrew
> file system, it was said to be placed in the public domain, but I have
> not heard more -- I would, nonetheless, be happy to obtain NFS as
> fast as possible.  I am ignorant of the details of NFS, I once read
> that it can use either tcp/ip or UDP, and while I realize that Myrinet
> can support tcp/ip I wonder if native use of the Myrinet for NFS would
> be faster.  To return to generalizations, a fast hard disk on a PC
> can give a peak of 160 MB and Myrinet can give a bandwidth on the
> order of 200 MB, this means that hard disks that are not local should
> not be a great bottleneck.  But this depends on the latency from
> various levels of handling the message, so I wonder if perhaps simple
> NFS has been "ported" to Myrinet to reduce the latency.
> Alan Scheinine  Email: [email protected]
> 

-- 
[                    Rob Latham <[email protected]>                  ]
[      Paralogic Inc. - www.plogic.com - www.xtreme-machines.com      ]