GFS to replace NFS... will this work?

Hey there,

  I currently have an elaborate NFS serving setup, and would like something 
faster and easier to manage without requiring the wholesale replacement of 
some fairly expensive hardware.

  Right now I have 4 600GB fileservers (ext3 on raid5 on mylex dac950 
controllers) which push data over gigE to a couple dozen front end boxes via 
NFS. The servers are mirrored in pairs (serv1a has the same data as serv2a, 
serv1b has the same data as serv2b), and one of the pairs is mounted on a 
given front end box. Initially we had 2 fileservers (for NFS performance and 
redundancy) but the dataset has outgrown the original 600gb and required an 
additional pair of servers.

  I'm looking ahead to a time when this next 600gb will be consumed and our 
bandwidth requirements grow yet again. It would be nice to not need to to grow 
into a 3 x 3 configuration, considering these servers, at the very cheapest, 
would run us between $10-15k USD, not to mention the organizational headaches 
it causes me.

  What I'm particularly looking for is a system which would satisfy these 
requirements, and ease my storage-related concerns for the forseeable future:

- Make use of current hardware with as few additional hardware requirements as 
  possible.

- Be as stable as the current setup. While the current schema is a little 
  complex, I've been able to get it to (or near) 5 nines availability. It 
  would be VERY BAD if a replacement system didn't match that availability.

- Could be used essentially as a replacement for NFS from an operational 
  standpoint (ie: that things would work pretty much as a plain ol' file 
  system from a daily operations perspective, like NFS currently does)

- Allow me to tack on servers as required to increase storage or performance.

  I've read through several of the GFS whitepapers, the FAQ, html howto and so 
on, but I'm not entirely clear on if my current hardware situation would lead 
to positive results. I'm under the impression that GFS using GNBD could be 
employed, but I've also seen comments about GNBD being unstable. I'm also not 
entirely certain on how easy it would be to tack on storage and make it 
available to the cluster (the white papers assure it can be done, but is a 
little light on the details of HOW).

  Has anyone else followed a similar migration path with experiences to share?


_______________________________________________
gfs-users mailing list
[email protected]
http://lists.sistina.com/mailman/listinfo/gfs-users
Read the GFS Howto:  http://www.sistina.com/gfs/Pages/howto.html