[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: FSFS performance on NAS/NFS

From: Justin Johnson <justinjohnson_at_gmail.com>
Date: 2007-03-08 16:25:41 CET

CollabNet said there wouldn't be any data integrity issues with
NAS/NFS and FSFS as long as only one server was accessing the
repository and the version of NFS supported locking. Mounting on
multiple servers concurrently and load balancing would require a
clustered file system.

My question was specifically related to performance though. Is there
anyone out there using FSFS repositories on NAS/NFS? Does the
performance "really really really suck?" Should I go with Berkeley
instead?

Thanks,
Justin

On 3/7/07, Rahul Bhargava <me@rahulbhargava.org> wrote:
> Well NFS locking can be tricky. NFSv3 protocol does not have built-in
> locking, it
> relies on another protocol NLM. Given the stateless nature of NFS
> protocol, NLM needs
> to store state on the server on who is locking what file range .
> Unfortunately there is no 'journal'
> on server-side to track that state during recovery. If the NFS server
> reboots it has to depend on
> a grace period for clients to come back and 'reclaim' their locks. If
> the NFS client fails to
> reach the server in the grace period (slow network etc), the server will
> wipe its state clean
> and essentially forget the locks that were held by the NFS client!
>
> In other words the recovery is not guaranteed to be 'safe' when clients
> and servers restart/fail.
> You may never encounter these issues but that does not mean you are safe
> under wacky
> failure scenarios.
>
> A single Apache server does not mean that there aren't multiple
> concurrent processes potentially accessing the
> NFS server.
>
> If you are using NFSv3 you may want to read this
> http://www.connectathon.org/talks06/talpey-cthon06-nsm.pdf on
> correctness issues with NLM. Not sure if NFSv4 bypasses all the issues.
>
> Les Mikesell wrote:
> > Rahul Bhargava wrote:
> >> Using NFS for any file system app like Subversion or CVS is
> >> dangerous. NFS clients typically cache file changes locally, that can
> >> cause weird errors (file size mismatches), the rename system call may
> >> not
> >> be atomic with NFS. If you are going to use a NAS array you are
> >> better off setting up iSCSI access rather than NFS.
> >
> > Note that client caching should only be a problem if you access
> > directly through the filesystem from multiple clients. If you mount
> > from a single server and use http or svnserve network acess from any
> > other clients everthing should work as long as nfs locks work on the
> > server.
> >
>
>
> --
> Rahul Bhargava
> http://www.rahulbhargava.org
> Phone: (925) 265-8801(W)|895-2201(M)
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: users-help@subversion.tigris.org
>
>

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Thu Mar 8 16:26:10 2007

This is an archived mail posted to the Subversion Users mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.