IBM SFS SAN Filesystem

Hi,

one of our customers in detroit has to deal with concurrent access from
different operating sytems to the same data structures (CAD/CAE and Lotus
Environment). We currently solved this problem with over 200 TB of NAS
Filers - but this didn't solve the problem of having the data in a lot of
different copies spread allover the company - and dealing with CIFS and NFS
Access concurrently is also a nightmare.

IBM told the customer of using SAN File System (Storage Tank) instead - and
of course IBM told that it will solve all problems....


The question therefo Has anyone expierences with IBMs SFS? What problems
did you encounter - or what problems would you see theoretically?

Thanks in advance

Matthias Ress

news.t-online.de wrote:
Hi,

one of our customers in detroit has to deal with concurrent access
from different operating sytems to the same data structures (CAD/CAE
and Lotus Environment). We currently solved this problem with over
200 TB of NAS Filers - but this didn't solve the problem of having
the data in a lot of different copies spread allover the company -
and dealing with CIFS and NFS Access concurrently is also a nightmare.

Why does a NAS solution lead to lots of different copies spread across the
company unless you mean duplicate copies on different NAS files. If that's
the problem then the answer is using larger NAS filers to consolidate.

IBM told the customer of using SAN File System (Storage Tank) instead
- and of course IBM told that it will solve all problems....


The question therefo Has anyone expierences with IBMs SFS? What
problems did you encounter - or what problems would you see
theoretically?


The biggest limitation at this time is the support for different OS
platforms vs. a NAS solution. CIFS/NFS based solutions are universally
supported without adding additional software to the host, whereas
StorageTank only works on a couple of OS platforms at this time.

The biggest question that IBM has yet to answer is how well will StorageTank
performance scale with multiple hosts accessing the shared device through a
lock a manager which can become a bottleneck.

As to practical expereince, there will be very little at this time and any
references that IBM provides will naturally be handpicked satisfied
customers.

Hi,

Why dont you take a look at SGI CXFS. As of today, it supports 5 operating
systems sharing the same filesystem without any data copies.
It has been in production for the past 5 years.

Cheers,
Arun

To detail the question:

Why does a NAS solution lead to lots of different copies spread across the
company unless you mean duplicate copies on different NAS files. If that's
the problem then the answer is using larger NAS filers to consolidate.


Problem 1:
Duplication of Data Structures - same data has to be copied for CAD via
NFS - Final Data is read for Transformation by CIFS - commercial information
is added again via NFS etc. etc.

Problem 2:
200 TB NAS Storage is really a lot - you end up with NAS Heads and SAN
Storage in the backend - but this is still a problem because Netapp supports
only specific SAN Environments, EMC² Celerra - same problem.

Problem 3:
Integrating HSM solutions and transfering older data to nearline or offline
environments. We started with SUN SAMFS - specially in the NFS Area. But
this caused a lot of other different problems (Performance, Interoperability
via Samba only, SUN specific solution, Backup/Restore...). But from what i
heard from IBM - Storage Tank wouldn't be a solution for HSM, would it?



IBM told the customer of using SAN File System (Storage Tank) instead
- and of course IBM told that it will solve all problems....


The question therefo Has anyone expierences with IBMs SFS? What
problems did you encounter - or what problems would you see
theoretically?


The biggest limitation at this time is the support for different OS
platforms vs. a NAS solution. CIFS/NFS based solutions are universally
supported without adding additional software to the host, whereas
StorageTank only works on a couple of OS platforms at this time.

The biggest question that IBM has yet to answer is how well will
StorageTank
performance scale with multiple hosts accessing the shared device through
a
lock a manager which can become a bottleneck.

As to practical expereince, there will be very little at this time and any
references that IBM provides will naturally be handpicked satisfied
customers.

Matthias Ress wrote:
To detail the question:

Why does a NAS solution lead to lots of different copies spread
across the company unless you mean duplicate copies on different NAS
files. If that's the problem then the answer is using larger NAS
filers to consolidate.


Problem 1:
Duplication of Data Structures - same data has to be copied for CAD
via NFS - Final Data is read for Transformation by CIFS - commercial
information is added again via NFS etc. etc.

So your NAS heads don't support CIFS and NFS simultaneously accessing the
same device? If all you platforms can access the NAS volumes why does data
have to be copied?


Problem 2:
200 TB NAS Storage is really a lot - you end up with NAS Heads and SAN
Storage in the backend - but this is still a problem because Netapp
supports only specific SAN Environments, EMC² Celerra - same problem.


True, but I doubt that IBM has anybody with anything remotely close to 200TB
of data in a StorageTank, frankly I'd doubt that they have a single instance
of StorageTank handling 20TB in a production environment. So you'd probably
end up with multiple StorageTanks to handle 200TB so it doesn't really
change the problem, just rearranges it ;-)

Problem 3:
Integrating HSM solutions and transfering older data to nearline or
offline environments. We started with SUN SAMFS - specially in the
NFS Area. But this caused a lot of other different problems
(Performance, Interoperability via Samba only, SUN specific solution,
Backup/Restore...). But from what i heard from IBM - Storage Tank
wouldn't be a solution for HSM, would it?

Nothing in the information produced by IBM to indicate support for HSM, so
I'm inclined to say no.

--
Nik Simpson

On Fri, 12 Dec 2003 11:03:21 -0600, Arun Ramakrishnan
wrote:

Hi,

Why dont you take a look at SGI CXFS. As of today, it supports 5 operating
systems sharing the same filesystem without any data copies.
It has been in production for the past 5 years.

CXFS 3.0 has introduced some nice new features (e.g. having options
for e.g. ssh controls).

.... And it works nicely with SGI's DMF HSM package.

Arun

Malc.

Hi Nik,

"Nik Simpson" schrieb im Newsbeitrag
...
Problem 1:
Duplication of Data Structures - same data has to be copied for CAD
via NFS - Final Data is read for Transformation by CIFS - commercial
information is added again via NFS etc. etc.

So your NAS heads don't support CIFS and NFS simultaneously accessing the
same device? If all you platforms can access the NAS volumes why does data
have to be copied?


Concurrent Access from NFS and CIFS is not really the problem - the problem
is dealing with different security Domains - mapping NFS Metainformation on
CIFS and vice versa. It doesn't matter which filer you use, but
Usermanagement is a nightmare whenever you use concurrent access with
thousands of users. The other problem is based on the fact, that you need
the same Data Structures (which doesn't mean that it's the same file).


Problem 2:
200 TB NAS Storage is really a lot - you end up with NAS Heads and SAN
Storage in the backend - but this is still a problem because Netapp
supports only specific SAN Environments, EMC² Celerra - same problem.


True, but I doubt that IBM has anybody with anything remotely close to
200TB
of data in a StorageTank, frankly I'd doubt that they have a single
instance
of StorageTank handling 20TB in a production environment. So you'd
probably
end up with multiple StorageTanks to handle 200TB so it doesn't really
change the problem, just rearranges it ;-)


Totally agree on that point ! Hope i can convince the customer too...

Problem 3:
Integrating HSM solutions and transfering older data to nearline or
offline environments. We started with SUN SAMFS - specially in the
NFS Area. But this caused a lot of other different problems
(Performance, Interoperability via Samba only, SUN specific solution,
Backup/Restore...). But from what i heard from IBM - Storage Tank
wouldn't be a solution for HSM, would it?

Nothing in the information produced by IBM to indicate support for HSM, so
I'm inclined to say no.


Again - totally agreement. I argue if it even would be possible to integrate
HSM in a specific Filesystem which is not part of the standard OS. I was
very convinced from the Avalon solution (EMC² nowadays) which integrated the
HSM-Filesystem-Standard (was it xdmp?) - think it makes more sense having
a standard in the existing filesystems that is used by HSM solutions than
having a HSM solution that is proprietary like SAMfs

Thanks

Matthias Ress

--
Nik Simpson