[OpenAFS] OpenAFS vs NFSv4?

[Ray Pasetes]

Rodney M Dyer wrote:
> 
> Umm, I understand you mean well here, but this is not the same.  If 
> there is no need for a client, then you just end up pointing all your 
> traditional clients CIFS/NFS to one network end-point.  The back-end 
> might be virtualized, but this solution doesn't distribute the load 
> across the network evenly.  Again, it's a silly NAS solution with a 
> "global namespace" back-end tacked on.  This solution would be fine for 
> web serving, but terrible for traditional client/server filesystem 
> access.  And, if you are going to use it for web serving, then there's 
> little point to having "global namespace".  This even supports my 
> conspiracy theory that NAS is just being pushed for business marketing 
> purposes, not technical merit.

Well, though not as clean as AFS, the clients don't have to connect to
a single network end-point. Any NAS head in the cluster can service the
the client, even if the information is not on that NAS head.  And, if 
the back-end storage is on a SAN, then any NAS head can cover for a 
failing NAS head. Basically they've put the "AFS client code" on the 
heads instead of the individual clients.

As for balancing load across the heads, one can use round-robbin DNS or
an L4 switch perhaps.  However, because they rely on NFSv3 or less, it
wouldn't necessarily be very good at WAN access.


-Ray Pasetes