[Gllug] distributed filesystems linux <-> solaris

[Ian Northeast]

Ben Fitzgerald wrote:
> Hi
> 
> Has anyone tried distributed filesystems between solaris
> and linux? We have had problems with large files when using
> nfs on linux with solaris.

Which way round?

I have seen problems with the server on Linux and the client on Solaris, 
admittedly this was a 2.2 kernel and the user space NFS server. The main 
symptom was that Solaris would create a file and immediately try to read 
it, to find it wasn't there. It would appear shortly afterwards. This 
was a serious problem when trying to compile things in NFS on Solaris.

With the server on Solaris (and indeed AIX) and the client on Linux, I 
have seen problems ranging from poor performance to total lockups unless 
the Linux defaults are overridden to "rsize=8192,wsize=8192". The man 
page recommends this anyway. With this setting I have had no trouble 
with either Solaris or AIX servers and Linux clients. The only total 
lockups were with the client on my laptop which runs Debian Woody with 
their stock 2.4.18 kernel and a rather slow PCMCIA NIC (actually it's 
the old PCMCIA bus which is slow). Setting the block sizes fixed it.

Linux does not have a good reputation as an NFS server. My feeling is 
that Solaris is better, and if you need to share files between the two 
it is best to put the server there. The same goes for AIX and probably 
other UNIXes.

I do have some Linux NFS servers with AIX clients in production without 
any problems but the access from the AIX side is very light so it 
doesn't prove much. In these cases the filesystems have been put on the 
Linux side because 99% of the access is from there. On all my other 
production heterogenous NFS setups the server is AIX or EMC's Celerra 
appliance fileserver and my fileserver at home is now Solaris. None of 
these give any trouble with the various Linux clients.

We are about to try a fairly heavyweight NFS exercise with a Linux 
server, which will be SLES8 with 2.4.21, and AIX clients (and some 
DC/OSX and Reliant UNIX thrown in, not to mention Samba and some Windows 
clients) in a disaster recovery test where we don't have our Celerra 
available (we have a DR Celerra but it's in a "synchronisation" mode 
unless we have an actual disaster and it can't serve files in this 
state). I wanted to use Solaris for this, but it's "contrary to policy" 
as Solaris isn't on our official list of OSs, even though the only 
people who will work on this temporary system are myself and someone who 
knows Solaris better than Linux. This should show whether my feeling is 
correct. The Celerra runs Linux BTW, but it doesn't use the normal Linux 
NFS code, EMC wrote their own. I'll report how this goes.

Regards, Ian

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug