clustering NAS
keane
Is it possible to cluster 2 NAS device using SUN Server,
If Yes, how is it going to be done??
Any suggestion will be appreciated.
thanks
Thomas H Jones II
Michael Vilain <mic...@NOSPAM.vilain.com> did thusly spew forth:
>In article <660be045.01080...@posting.google.com>,
> ts...@pacific.net.sg (keane) wrote:
>
>> Is it possible to cluster 2 NAS device using SUN Server,
>> If Yes, how is it going to be done??
this question is unclear. a NAS is typically going to be some sort of
NFS or SMB type solution. you the protocols really arent set up for
forwarding an NFS or SMB connection from a NAS device, through a host
to another (third) system.
If, on the other hand, the question is, is it possible to set up a
clustered NFS server, then the answer is yes.
>At first blush, I would say very strongly NO. Veritas and SUNcluster
>software relies on having access to the underlying physical I/O to
>access specific blocks. Solaris keeps track of the filesystem structure
>through the layer of the cluster "middleware" which decides where that
>data recideds (e.g. what disk connected to which SCSI or Fibre
>controller). Because Solaris is doing the dirty work, additional
>distributed software to manage coordination and file locking is also
>implemented at the filesystem level. This allows multiple Solaris
>systems to write to a single device which is physically attached to each
>of those systems.
Or, you can configure a high-availabilty NFS service within your cluster
configuration. It answers to the shared IP. Assuming youve set it up
correctly (matching majors/minors for all the filehandles to access),
your clients will be able to follow the NFS service, no matter which host
it runs from. if you dont want to delint all the details, let automounter
do the work for you.
>This is not the case for Network Appliances NAS boxes. They have their
>own OS that serves NFS over a wire, which is not capable sharing a
>filesystem as if it were mounted locally. NFS allows for shared write
>access, but each client system sees the filesystem as remote. This is
>not clustering. The concept of sharing disks in a NAS is bogus as the
>entire box serves a single NFS volume.
It should be noted, however, that a NetApp array can be set up in a
high-availability mode. Typically, there are two storage controller
heads attached to all of the shelves in the array. under normal conditions,
each head acts and serves independently. in the event of a failure of
on of the controller heads, the surviving head impersonates the dead one
and continues to handle the requests sent for its services.
>However, if you want to cluster systems to provide failover, that's
>built into NFS. Just specify multiple IP (one for each interface on
>separate nets in the NAS box) in the dfstab file or automounter map.
>It's supposed to failover to the other address. I don't think the NAS
>box can failover network paths transparently (unless they've updated
>their OS).
Which is fine, if you have replicated file store. If youre already using
a clustered storage system, the high availability NFS modules are usually
fairly trivial to implement.
-tom
--
"You can only be -so- accurate with a claw-hammer." --me
Robert Milkowski
> It should be noted, however, that a NetApp array can be set up in a
> high-availability mode. Typically, there are two storage controller
> heads attached to all of the shelves in the array. under normal conditions,
> each head acts and serves independently. in the event of a failure of
> on of the controller heads, the surviving head impersonates the dead one
> and continues to handle the requests sent for its services.
but NetApp does not support network failover in HA configuration, AFAIK.
Some filers (Clariion I mean) do.
--
Robert Milkowski
mi...@wp-sa.pl
Calle Dybedahl
> but NetApp does not support network failover in HA configuration, AFAIK.
My NetApps do. If they didn't, what would the point of the failover be?
In fact, NetApp's failover is more fundamental than any other I've
ever seen, since the secondary will take over not only the IP but also
the MAC address of the failed primary. The drawback is that the
machines in the pair must be physically very close.
--
Calle Dybedahl | UNIX-admin | Telenordia Internet | c...@algonet.se
Robert Milkowski
>>>>>> "Robert" == Robert Milkowski <rmilk...@wp-sa.pl> writes:
>
>> but NetApp does not support network failover in HA configuration, AFAIK.
>
> My NetApps do. If they didn't, what would the point of the failover be?
in case of hardware probles with cotrolers, etc.
> In fact, NetApp's failover is more fundamental than any other I've
> ever seen, since the secondary will take over not only the IP but also
> the MAC address of the failed primary. The drawback is that the
> machines in the pair must be physically very close.
Hmmm, are you sure? Last time I've talked with technical/sales man from NetApp
he said there's no network failover (we were talking mainly about F840C, but
it doesn't matter). What system version are you talking about?
--
Robert Milkowski
mi...@wp-sa.pl
Calle Dybedahl
> Hmmm, are you sure? Last time I've talked with technical/sales man
> from NetApp he said there's no network failover
I'm quite certain, since I've had a live-fire test when a PSU gave
out. Not only did it failover, it did it so fast that the clients
didn't even display "NFS server foo not responding" messages. Failing
back once the PSU had been replaced took longer, something like 15
seconds.
Each machine in the pair has an extra NIC, configured identically with
the live NIC in the partner machine but normally not active.
> (we were talking mainly about F840C, but it doesn't matter). What
> system version are you talking about?
I've got a pair of F630s, and I honestly don't remember which version
of OnTap is on them. I haven't had to touch them since January last year.