You are close but are missing some key concepts with SQL clustering. There
are two distinct components involved, nodes and instances. Nodes are the
physical machines that comprise the cluster. Instances are the virtual SQL
servers. A node can host zero or more instances. If you set up a four node
cluster, you can park an instance on each of the three nodes with the fourth
node designated as the first failover node for each instance. As far as the
cluster is concerned, all host nodes are equal so you can allocate the
instances amongst the nodes as you see fit. From the client perspective,
all interaction is with the virtual server (instance) so you don't care
where the instance is actually hosted.
As for storage, you need a storage mechanism that is accessable to all
nodes, typicaly a SAN. The cluster service arbitrates ownership between the
specific nodes so all the resources necessary for any single instance to
function are always together on a single host node. The instance IP address
is one of the unique resources within the virtual server resource group.
As with all service packs, SP4 is cluster-aware, so that if you run the
installer from the node currently hosting an instance, it upgrades the local
binaries on all nodes for that particular instance. You will have to run
the Service Pack separately for each instance, however it is perfectly fine
to run instances at different SP and hotfix levels within a cluster.
FYI, I have built and managed a four-node, three instance cluster like you
have described, up through SP3a + hotfix 9??. I haven't tried SP4 yet but I
don't see any particular problems except for the already resolved AWE issue.
--
Geoff N. Hiten
Senior Database Administrator
Microsoft SQL Server MVP
[quoted text, click to view] <bcrenshaw99@yahoo.com> wrote in message
news:1124814061.497589.15020@g49g2000cwa.googlegroups.com...
> We are looking at setting up the following cluster:
>
> 3 Active nodes:
> Each server needs to be completely distinct with its own unique
> database and its own virtual server and ip. These nodes do not share
> the same storage with the other active nodes. Active nodes must only
> process for their own db's on the server itself
>
> 1 Passive node:
> Common backup node for all the active nodes. Cluster service must be
> able to transfer the disk ownership from the failed active node to this
> one.
>
>
> Is this possible with Windows 2003 and SQL Server SP4? I'm a little
> concerned due to the way you need to apply sp4 to a SQL cluster.
>
> Any advice would be much appreciated
>
> Bill
>
