Clustering more than one SQL Server instance at a time

Hi, ever since last week I have a doubt about a type of clustering
implementation. Is it possible to cluster more than one SQL Server
instance. I mean i know it is possible to cluster one instance... so
that when that instance fails the other one enter into context be it
either manually or automatically... but what happens when there are 3
instances? if one instance fails the other two instances would be
forced to pass their execution to the other server even when maybe
everthing was ok with them
Thanks in advance for the response
AlanI think you misunderstand the relationship between an instance and a server
in a cluster.
A cluster is a group of machines that share a common storage unit and are
configured for failover. The cluster software arbitrates the shared storage
unit so that only one host computer (node) can control a logical disk (LUN)
at a time. Note that the storage logical disk appears to a node as a
physical disk.
A node is a particular host computer within a cluster.
An instance is a virtualized SQL server. It has one or more LUNs from the
shared unit, its own network name and IP address. It is hosted on a single
node at a tim, but can move to another node either in response to a failure
event or by administrator command. It is important to understand the
separation of nodes and instances. Instances are part of the overall
cluster. Nodes are either current or potential hosts. This type of
clustering is called "shared nothing" since no node requires any resource
from any other node in order to operate. When an instance moves to another
host node, the client sees the SQL Server stop and a restart. It still
connects exactly as before, since the IP address and network name moved to
the new host node along with the rest of the instance resources.
Just as a stand-alone system can support multiple instances, o can a
cluster. Each instance is completely independent of any other instance
(more of the "shared nothing" design). Each new instance must have its own
disk(s), IP address, and network name. You can move instances around on
various cluster nodes as you want, completely independent of each other. Of
course, each host node has a finite amount of processor and memory, which
must be shared between any instances it currently hosts.
So yes, you can have many instances in a cluster. Note that with shared
storage systems other than SCSI, you can have more than two nodes in a
cluster. Clusters with X nodes and X- instances are classified as N-1
Clusters.
--
Geoff N. Hiten
Senior SQL Infrastructure Consultant
Microsoft SQL Server MVP
<aferrandiz@.gmail.com> wrote in message
news:1187730948.651429.86880@.r23g2000prd.googlegroups.com...
> Hi, ever since last week I have a doubt about a type of clustering
> implementation. Is it possible to cluster more than one SQL Server
> instance. I mean i know it is possible to cluster one instance... so
> that when that instance fails the other one enter into context be it
> either manually or automatically... but what happens when there are 3
> instances? if one instance fails the other two instances would be
> forced to pass their execution to the other server even when maybe
> everthing was ok with them
> Thanks in advance for the response
> Alan
>|||Hi Geoff,
I believe that the cluster appears to the application as a single IP
address that it connects to. Is it mandatory that this IP address
happens to be one of the nodes participating in the cluster? Or is it a
virtual one?
If yes, what happens when the node whose IP address is visible to the
application happens to crash? If no, does the externally visible IP
address that was mapped to the node that crashed get transferred to
another node that is available and has been elected as the new interface
to the application?
How does this happen exactly?
Thanks,
Sanchet|||The IP ia a virtual one and cannot be a host IP address.
Geoff N. Hiten
Senior SQL Infrastructure Consultant
Microsoft SQL Server MVP
"Sanchet Dighe" <sanchetd@.gmail.com> wrote in message
news:46CC35EA.3030706@.gmail.com...
> Hi Geoff,
> I believe that the cluster appears to the application as a single IP
> address that it connects to. Is it mandatory that this IP address happens
> to be one of the nodes participating in the cluster? Or is it a virtual
> one?
> If yes, what happens when the node whose IP address is visible to the
> application happens to crash? If no, does the externally visible IP
> address that was mapped to the node that crashed get transferred to
> another node that is available and has been elected as the new interface
> to the application?
> How does this happen exactly?
> Thanks,
> Sanchet|||Hi, ever since last week I have a doubt about a type of clustering
implementation.
Is it possible to cluster more than one SQL Server instance. I mean i know
it is possible to cluster one instance...
-- Yes is possible, for each SQL Server instance in a custer you need the
following :
-- A group for each instance of SQL
-- In the new group one disk(lun) o more, It is for install the new SQL
instance, if you plan to have more than 1 disk for the instance at the end
you need to add the disk like dependecies of SQL Server resource in the group
-- Ip Addrees, the setup ask for the ip and at the end of the setup
automatically add the ip resource into the group,
-- SQL name (instance name) , the setup ask for the instance name and at
the end of the setup automatically add the instance name resource into the
group,
so that when that instance fails the other one enter into context be it
either manually or automatically...
-- automatically
but what happens when there are 3 instances? if one instance fails the other
two instances would be forced to pass their execution to the other server
even when maybe everthing was ok with them
-- No, just the instance failed pass to other node in the cluster.. the
other instances still working without problem
Let me know any doubt about the problem...
"aferrandiz@.gmail.com" wrote:
> Hi, ever since last week I have a doubt about a type of clustering
> implementation. Is it possible to cluster more than one SQL Server
> instance. I mean i know it is possible to cluster one instance... so
> that when that instance fails the other one enter into context be it
> either manually or automatically... but what happens when there are 3
> instances? if one instance fails the other two instances would be
> forced to pass their execution to the other server even when maybe
> everthing was ok with them
> Thanks in advance for the response
> Alan
>|||> If yes, what happens when the node whose IP address is visible to the
> application happens to crash? If no, does the externally visible IP
> address that was mapped to the node that crashed get transferred to
> another node that is available and has been elected as the new interface
> to the application?
> How does this happen exactly?
How does this work is how TCP/IP ARP (address resolution protocol) works.
When node A that currently has an IP resource crashes and the cluster service
decides to bring the IP resource up on node B, node B will broadcast an ARP
request to force all the nodes on the subnet to update their ARP cache to map
the IP address to the machine address of node B. So when an external computer
tries to access that same IP address, that IP address now will be resolved to
node B.
I believe in Longhorn this will work slightly differently because the "same
subnet" restriction will be lifted. But the general principle is the same. So
the IP address is not really 'moved', it is just being resolved to a
different computer.
Linchi
"Sanchet Dighe" wrote:
> Hi Geoff,
> I believe that the cluster appears to the application as a single IP
> address that it connects to. Is it mandatory that this IP address
> happens to be one of the nodes participating in the cluster? Or is it a
> virtual one?
> If yes, what happens when the node whose IP address is visible to the
> application happens to crash? If no, does the externally visible IP
> address that was mapped to the node that crashed get transferred to
> another node that is available and has been elected as the new interface
> to the application?
> How does this happen exactly?
> Thanks,
> Sanchet
>