Ok, I've got the cluster setup and running, but having never done this,
I'm not sure if I'm setting up SQL right... We're trying to migrate
our multitude of SQL Server running on older hardware to the new
cluster, but I want to make sure we don't shoot ourselves in the foot.
Here is what we've got:

Specs:
3 HP BL20P Blade Server (Twin 3.6GHz Xeon, 4GB Ram)
1 HP MSA1000 w/ Twin Fibre Switches (Dual Path Redundancy)

Current Setup:
Windows 2003 Enterprise, 20GB C:, 10GB D: (Pagefile), 37GB E: Data
MSA1000 is currently configured with 4 36GB Arrays (Quorum, 2 for Trans
Logs, and 1 for Backups) and 2 120GB Arrays (Database Data), but there
is about 1.2TB left on the controller for additional space.

Followed all the instructions, Public IPs, Private IPs, etc...

Installed SQL on the first two nodes (SQLCL01 & SQLCL02) and have two
Virtual Servers (Same name, actual server name is longer and unique),
and then two instances, INST1 on SQLCL01, and INST2 on SQLCL02.
SQLCL03 is the failover server which of course is identical to the
first two. We're never expecting to have 2 fail, but we may add a 4th
Server/Node in the future (we have 5 additional slots in our two Blade
Chassis).

[quoted text, click to view]
cluster, but obviously I don't know if #1 I should try and install more
than 2 instances, or #2 if I even can. I've tried rerunning the SQL
Setup just to see, and all it lets me do it "modify" the current
install or remove it, it won't let me add another instance.

So... What am I looking at here? Is this the optimum configuration
for now, or can I do more? What about memory? Should I limit each SQL
Server instance to a certain level of RAM, say 3GB? Or possibly 2GB,
incase both the two main servers ever fail and everything gets forced
to the 3rd? These servers will pretty much only be used for SQL,
nothing else, so there isnt' too much worry about applications battling
for memory.

Thanks in advance. I know some of these questions may seem rather
newbish, but I've installed and administered SQL2k before, but never a
cluster... so this is new ground for me and the documentation out
there is not very helpful, most of it refers to SQL2k on Windows 2000,
not Windows Server 2003.

Jon Casimir

Lotsa comments inline.

[quoted text, click to view]
I am not a big fan of blade servers as cluster nodes. Too many single
failure points for what is intended to be a highly available system.

A well-tuned, dedicated SQL server should have minimal need for a paging
file. If you are paging heavily, you have something tuned wrong.

Backups should never be stored on the same host computer or storage array as
the primary data store, even if they are on separate physical disks. Backup
across the net to a file share for immediate use and archive those files to
tape for longer retention periods.

[quoted text, click to view]

This doesn't sound right. You should have each Virtual Server\Instance
combination installed on all nodes on the cluster so you can fail over as
needed. During install time you can select which cluster nodes to install
SQL to. You can set the preferred node order of each Virtual Server later
independently so they start and fail where you choose.

On a multi-node, multi-instance cluster, I usually only worry about
first-order failures. If I have more than one instance go south on me, it
is usually the entire cluster that bombs. If I have one instance with a
problem, somebody competent better be standing in front of it fixing the
problem within 30 minutes. You can adjust memory settings and failover
order at that time.

[quoted text, click to view]
SQL won't let you add a new virtual server unless there is at least one
unassigned cluster disk resource to anchor the instance. Whether you
"should" install more instances is another matter. Each instance looks and
acts like a separate server on the network. Generally, multiple instances
in a cluster are used to manage security and performance. In my history, I
find that one instance per node + one spare is an optimal configuration, but
your needs with this server consolidation project may vary.

[quoted text, click to view]

Since you have paid for Enterprise Edition anyway you should max out the
memory, although your choice of blade servers as hosts may limit that
expansion capability. You will need to consider what happens during a
failover so that you can tolerate "stacking" multiple instances on the same
nost node.

[quoted text, click to view]

Most of the considerations for Windows 2000 clustering apply to Windows
2003, except for some installation gotchas. Unless you are sure something
from Windows 2000 doesn't apply, assume it does.

Now is the best time to ask "dumb" questions. Later, when your "highly
available" database solution that you bet your job on is down is the worst
time.

[quoted text, click to view]

Geoff N. Hiten
Senior Database Administrator
Microsoft SQL Server MVP