[NFS] limit on filesystem size?
Trond Myklebust
> Hello, we are in the planning process for the setup of a large
> (~ 100 TB) NFS server. My understanding is that there is a
> limit of 2TB for filesystem sizes for local (!) filesystems
> under Linux (at least for ext2/ext3; I am not sure about xfs).
> My question really is: is there a limit for the filesystem size
> that a Linux NFS client (2.4.18 or the upcoming 2.4.19 kernel)
> can mount from a (possibly non-Linux) NFS server? E.g., can I
> mount a 20TB filesystem under Linux?
> (I've asked this question on the beowulf mailing list and the
> answer was no. Now I ask the experts - hoping for a different
> answer :-)
If you can mount it, then you can export it...
For NFSv2 there is a 32-bit size limit on the files you can export
whereas for NFSv3, that limit is 64-bits, but those limits are
*per-file*.
Otherwise, you might run into problems when you try to run 'df' in
order to read filesystem statistics (NFSv2, the limit is 41-bit
- NFSv3 the limit is 64-bit) but that shouldn't really be a critical
feature.
Cheers,
Trond
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - N...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs
Ragnar Kjørstad
> How are you planning on creating this filesystem? I have tried
> creating 2TB filessystems here for a while and have had little or no
> luck w/ Solaris or linux. Using Disksuite for solaris and LVM and MD for
> linux no luck. I have not tried veritas VM yet but even calling up SUN
> the answer is that 1TB is the limit per filesystem currently under solaris 8
> & 9
>
> Linux is really around 980GB or so where it starts to flake out.
Linux goes up to 2 TB if you use a scsi-driver that use unsigned
sector-numbers and a recent version of LVM (or no LVM).
--
Ragnar Kjørstad
Big Storage
Trond Myklebust
> Please, bear with me - I am by no means an expert on these
> issues. But my impression was that the problem is in
> include/linux/blkdev.h
> struct request { ...
> unsigned long sector;
> with results in a (local) filesystem size limit of 2TB (unless
> sectors larger than 512 bytes are used).
> Am I understanding correctly that you are telling me that this
> does not apply to NFS and that despite this definition I could
> even reexport (user space NFS?) such a filesystem?
I repeat: yes!
NFS does not know anything about sectors. It indexes *files* by means
of an opaque 32-byte (NFSv2) or 64-byte (v3) filehandle. The latter
can be considered to be the equivalent of 'file descriptors' as used
by libc does in open()/read()/write().
Rexporting filesystems is never a good idea though. Better to allow
all NFS clients direct access to the server...
Cheers,
Trond