I am trying to install a cluster of two computers. Both
of them are going to be production machines which have
similar applications installed on it. Both have SQL
installed and are database servers. As data comes into
one of them the SQL on this first machine replicates the
data for every transaction to the 2nd machine. Thus both
have similar data at almost all times.

On failure I want the cluster service to make the 2nd
computer as the new server where the data can be inserted
automatically until the first server is brought to action
again.

The clustering architecture in windows wants the 2
machines to not have dat on them but on an external SCSI
hard drive. But this sturcture fails to provide any
service when the external hard drive itself fails. Our
approach takes care of hard drive failure as well a
machine failure.

How do I go about doing the cluster service installation
as the service, while installation, prompts that no
seperate SCSI bus found (during the installation of the
first node)

Thanks in advance.

Shailesh

Microsoft's implementation of clustering uses a shared set of drives to
store the data (SAN). What you want to do is network load balancing with
only 1 node being active at a time.

If you are worried about the shared external drives and their associated
equipment failing, add another SAN with separate IO controllers.

It is not recommended that you run a SAN on SCSI, but rather on Fiber
Channel.

A possibility is always buying a Unisys ES 7000 and clustering over the
virtual machines.

Replication always has a time lag, so what are your tolerances for data loss
(1, 5 or 10 minutes) ?

Regards
--------------------------------
Mike Epprecht, Microsoft SQL Server MVP
Epprecht Consulting (PTY) LTD
Johannesburg, South Africa
Mobile: +27-82-552-0268
IM: mike (AT) NOSPAMepprecht (DOT) net

Specialist SQL Server Solutions and Consulting

"Shailesh" <shailesh (AT) softhome (DOT) net> wrote

Hello Mike,

I shall be using transactional replication which wont be
time consuming at all. So I do not want the clustering
service to handle replication.

The two nodes will have their own hard drives and their
own version of SQL Server running. The data on them will
be nearly simlar because of the replication performed by
SQL Server on machine 1.

On a failure of machine 1, I simply want the clustering
service to hand over the control to Machine 2. No
replication of data is performed by this server. At all
times the nodes in the cluster will be handling their own
data (not the shared drive).

Is there such a design possible?

Thanks,

Shailesh

Yes. it is possible

How will this be possible? Windows Advance Server does
not allow me to install the cluster because it says no
shared SCSI Bus found. Lets say I install this SCSI
drive, how will I install the second node of the cluster?
I do not want the second cluster to use the SCSI drive
used by the first machine.

Thanks
Shailesh

Well, anything is possible. But the question is whether
it's best practice.

Unless you have a very simple database that rarely changes
and you can afford to lose data, the SQL2000 replication
is not a good method for the kind of disaster recovery you
are after.

Also, the Microsoft Cluster Service is not built to handle
the SQL Server failover described in your scenario. The
MSCS clustered SQL Server requires shared storage for
database/log files.

Linchi

Mike -

If you have two computers not connected, they are not
clustered. SQL Server 2000's implementation has a shared
disk mode, so each instance would "own" its own disks, and
that instance's resources (including disk) would be owned
by one node at any given time. They cannot be shared.

In clustering, if one instance fails, it will move lock,
stock, and barrel to the other node. It goes through a
stop and a restart, so I'm not sure why you are using
replication.

Allan

If possible, how so? During cluster installation, windows
says that it hasnt found a seperate SCSI bus on shared
SCSI drive. Even if I do install an second SCSI hard
drive, I do not want the second machine to use this shared
drive for any purpose. I want the second node to use its
own seperate drive. So is how do I achieve this design
with no shared drive involved at all?

Thanks,
Shailesh

But you can have clustering installed which is without a
shared disk but has a seperate private ethernet node-to-
node connection in additon to the public ethernet one.

My design has no need for the shared disk as the data
will reside on individual machines.

Thanks
Shailesh

It's important to be clear about exactly what clustering
you are talking about. From your other posts in this
thread, it looks like you are talking about server
clustering using Windows 2000 Advanced Server. If that's
the case, you must have a shared cluster disk(s).

Then, Windows 2000 AS server clustering may not be the
technology for your design.

Linchi