NFS tuning over ZFS

George Mamalakis mamalos at eng.auth.gr
Tue Mar 29 10:13:12 UTC 2011


Hi everybody,

I want to use a setup of nfsv3,krb5i mountpoints for my users' 
homefolders. The clients are linux (Ubuntu) and the server is 
FreeBSD-8-STABLE. The clients use autofs for mounting the users' 
homefolders. Once a user logs in, their folder is mounted, and the user 
has access to their homefolder via nfs. When I run ls(1), the system 
sometimes "stalls" before it responds. At the moment I am in the pilot 
phase, so there is no load on the server or client to cause this stall, 
since I am the only user using this setup. The delay in this ls(1) 
command is present even if I mount the homefolder without the use of 
kerberos.

Hence, I started reading about NFS tuning both for linux and FreeBSD. 
What I found in a few resources regarding linux, was to enable async 
mounts on the client if I was willing to take the risk of compromising 
data integrity. I've tried mounting the share with the async option set, 
just for the sake of experimentation, but with no luck (mount(8) doesn't 
show 'async' among the mount options). I didn't look into it more 
deeply, so I moved on.

As far as my next tuning candidate is concerned -the server- the shares 
are stored on a ufs partition of an esxi FreeBSD-image, for the moment. 
Once my proof-of-concept period is completed, the setup will be migrated 
to a real system and the homefolders are meant to be stored on a ZFS 
zraid partition. The server uses a 6-disk raid on an IBM 8k raid 
controller with memory and battery backup. I will create a 15GB volume 
(raid6) for the root filesystem, and will "export" the remaining space 
of each disk as a separate, equal-size volume. This way, I will be able 
to create the ZFS raid over these 6 volumes and then I will export the 
filesystem using NFS.

 From what I've read so far, both on this list and on 
http://wiki.freebsd.org/ZFSTuningGuide, it is stated that if ZIL is 
disabled or put on a SSD separate disk, NFS over ZFS performance is 
enhanced. On some mails of this list, specifically those with the 
subject ""zfs, nfs and zil" started by Claus Guttesen, it is mentioned 
that ZIL 'may' be disabled if the underlying system uses an analogous 
data integrity mechanism.

So, after all this intro, here are my two simple questions:

1) If I enable this mechanism on my controller, will my filesystem be 
safe in a case of power outage?
2) Is there a good, up-to-date guide for NFS performance for FreeBSD?

Thank you all for your time in advance,

mamalos

-- 
George Mamalakis

IT Officer
Electrical and Computer Engineer (Aristotle Un. of Thessaloniki),
MSc (Imperial College of London)

Department of Electrical and Computer Engineering
Faculty of Engineering
Aristotle University of Thessaloniki

phone number : +30 (2310) 994379



More information about the freebsd-stable mailing list