Carsten Aulbert wrote:
Carlos O'Donell wrote:If you run diskless nodes then distributed filesystems are not going tovery usefull ;) I have not seen any serious benchmarks for all[1] of these filesystems. Each is largely effected by the topology and technology of your cluster.[1] CODA, OpenAFS, GFS, OpenGFS, Lustre (Am I missing any?).Speaking of it, does any of you know if any of these file systems are reliable against failure of a single (or a few nodes) out of say 150?How about automatic rebuild and such things?
Hmm, well, it could probably work well if you don't have a lot of data (then you don't need to create a lot of partition/volumes). I found their (OpenAFS/Coda) limited sizes a major problem. They're not exactly the speediest things I've layed my hands upon either... Intermezzo seems more promising - but is still to unmature for production settings.
I'd like to gather any information possible since we're in the process of building a cluster with ~ 150 nodes and about 100-140 GB/node of 'free' hard drive space.
To large - to large :-\ -- David J. M. Karlsen http://www.davidkarlsen.com - http://mp3.davidkarlsen.com +47 90 68 22 43