Very Computer

by **Paul Arche** » Sun, 13 Jun 2004 11:34:02

I'm working with a company that has a Solaris (8, I think) NFS server
sharing most of a terabyte worth of very small files (1-2k each).
There are about 100 Linux clients using the data, which is read-mostly.

I'm investigating solutions for implementing some load balancing and
redundancy. So far, the best thing I've seen is CODA/AFS, but I suspect
that's primarily because I don't have enough exposure to large systems.
(I've been a Sun instructor for some time, but haven't done anything with
the storage side of the house.)

So, can anyone give me some suggestions? I have the feeling that a SAN might
be a solution, or part of a solution, but I don't know if it would be feasible
(or even possible) to hook all 100 clients up to a SAN, and I have read a few
articles/papers/sites that suggest that clustering NFS is made difficult by
the fact that different servers will have different filehandles for the same
basic datasets, unless care is taken to make the filesystems identical.

TIA,

Paul Archer

by **Greg Menk** » Sun, 13 Jun 2004 21:39:08

> I'm working with a company that has a Solaris (8, I think) NFS server
> sharing most of a terabyte worth of very small files (1-2k each).
> There are about 100 Linux clients using the data, which is read-mostly.

> I'm investigating solutions for implementing some load balancing and
> redundancy. So far, the best thing I've seen is CODA/AFS, but I suspect
> that's primarily because I don't have enough exposure to large systems.
> (I've been a Sun instructor for some time, but haven't done anything with
> the storage side of the house.)

> So, can anyone give me some suggestions? I have the feeling that a SAN might
> be a solution, or part of a solution, but I don't know if it would be feasible
> (or even possible) to hook all 100 clients up to a SAN, and I have read a few
> articles/papers/sites that suggest that clustering NFS is made difficult by
> the fact that different servers will have different filehandles for the same
> basic datasets, unless care is taken to make the filesystems identical.

I guess the first question is do you know you need to load balance-
meaning what measurements have you taken wrt network utilization at
the server & the disk I/O thats going on?

Gregm

by **Paul remove-the-nospam Da** » Sun, 13 Jun 2004 23:11:58

> I'm working with a company that has a Solaris (8, I think) NFS server
> sharing most of a terabyte worth of very small files (1-2k each).
> There are about 100 Linux clients using the data, which is read-mostly.

If they're mainly read-only, this is a perfect place for cachefs. But
then the Linux guys don't seem to have put cachefs into Linux yet...
coda, AFS etc have client-side caching, but then you require your
Solaris file-server to be speaking coda/afs - which doesn't look easy.

Quote:> So, can anyone give me some suggestions? I have the feeling that a SAN might
> be a solution, or part of a solution, but I don't know if it would be feasible
> (or even possible) to hook all 100 clients up to a SAN, and I have read a few
> articles/papers/sites that suggest that clustering NFS is made difficult by
> the fact that different servers will have different filehandles for the same
> basic datasets, unless care is taken to make the filesystems identical.

Yeah, clustering rw NFS wouldn't be pretty...

My immediate suggestions would be:
- Convert the Linux guys to Solaris 9 x86 with all the gnu utils to make
them feel at home, and then use cachefs.
- Upgrade the Sun box's hardware.

Not very helpful suggestions though...

PD
--
Paul Day Web: www.bur.st/~paul GPG Key ID: 2EF4ED23

by **Atro Tossavaine** » Tue, 15 Jun 2004 21:40:11

> I'm working with a company that has a Solaris (8, I think) NFS server
> sharing most of a terabyte worth of very small files (1-2k each).
> There are about 100 Linux clients using the data, which is read-mostly.

> I'm investigating solutions for implementing some load balancing and
> redundancy. So far, the best thing I've seen is CODA/AFS,

It is very likely that the best thing for your requirements _is_ AFS.

Signed, AFS administrator.

--
Atro Tossavainen (Mr.) / The Institute of Biotechnology at
Systems Analyst, Techno-Amish & / the University of Helsinki, Finland,
+358-9-19158939 UNIX Dinosaur / employs me, but my opinions are my own.
< URL : http : / / www . helsinki . fi / %7E atossava / > NO FILE ATTACHMENTS

by **Atro Tossavaine** » Tue, 15 Jun 2004 21:40:47

Quote:> If they're mainly read-only, this is a perfect place for cachefs. But
> then the Linux guys don't seem to have put cachefs into Linux yet...
> coda, AFS etc have client-side caching, but then you require your
> Solaris file-server to be speaking coda/afs - which doesn't look easy.

It's not all that difficult, really.

AFS rocks.

--
Atro Tossavainen (Mr.) / The Institute of Biotechnology at
Systems Analyst, Techno-Amish & / the University of Helsinki, Finland,
+358-9-19158939 UNIX Dinosaur / employs me, but my opinions are my own.
< URL : http : / / www . helsinki . fi / %7E atossava / > NO FILE ATTACHMENTS

Very Computer

NFS load balancing advice needed

NFS load balancing advice needed

NFS load balancing advice needed

NFS load balancing advice needed

NFS load balancing advice needed

NFS load balancing advice needed