Re: [nfsv4] Global namespace for v4.1?

brent.callaghan@sun.com

Message-ID: <3FE4CE6A.3000104@sun.com>
From: Brent Callaghan <brent.callaghan@sun.com>
Subject: Re: [nfsv4] Global namespace for v4.1?
Date: Sat, 20 Dec 2003 14:34:18 -0800

I also prefer a single method of querying the namespace,
that is client-based (David's option #1).

An important feature of a client-based solution is that
it's accessible to clients that are connecting to existing NFSv2/v3
servers.  The same service could also extend the namespace
to other file access protocols.

A global namespace that's not tied to a specific version of
the NFS protocol is very much easier to deploy in a
customer network with a variety of NFS clients and servers.

Why wait for protocol changes and/or NFSv4 deployment ?
Why not do it now ?

	Brent

Jon Haswell wrote:
> David.Robinson wrote :
> 
> 
>>If we have standardized the namespace information, then
>>it is relatively simple for a smart client to directly
>>query it and skip the need to use the server for a
>>referral.  This may help environments where there are
>>many referrals or the NFS latency is high.
> 
> 
> But should we allow clients two methods for obtaining this information ? I
> would argue that providing two ways of doing things, unless there is a very
> strong reason to do so, just increases complexity and testing. Additionally
> when we get into global namespaces with migration and replication we need
> to ensure that clients are redirected around the global namespace in sync
> with the movements of data, if we allow clients to bypass the server and go
> straight to a separate namespace server we have to keep more entities in
> sync, and yes it's possible but more complex with more testing required .
> 
> I think we should support simple clients who deal only with the NFS server,
> keeping their installation and operation simple.  In the case where we have
> many referrals, to the point where a server that is not serving data but
> only referals is a bottleneck isn't the answer to replicate this server ?
> If NFS server latency is high and the server itself isn't the bottleneck
> isn't it also reasonable to assume the client to namespace server latency
> will also be high, i.e. we are dealing with a client with a poor network
> link to the rest of the world (and it would be surprising if the NFS server
> or namespace server were on such a poor communication link).
> 
> Jon
> 
> 
> _______________________________________________
> nfsv4 mailing list
> nfsv4@ietf.org
> https://www1.ietf.org/mailman/listinfo/nfsv4

_______________________________________________
nfsv4 mailing list
nfsv4@ietf.org
https://www1.ietf.org/mailman/listinfo/nfsv4