I have a wild prediction. Within 5 years all business servers will be using clustering. What do Terminal Services, wireless networks, biometric logon and
clustering, all have in common? The answer is that at first, I thought
that each was rubbish. However, gradually, in each case I have done an about turn
and now I am a staunch advocate for all these technologies. I am
particularly excited by clustering for Exchange 2003.
The master plan is to deploy clustering and so eliminate a single
point of failure. If, for what ever reason, an Exchange server
is not available, then another server is firstly aware, and secondly has the
capability of taking over the dead servers role. The email will
continue to flow!
Clustering neatly illustrates how a new Microsoft technology evolves.
In the 1990's Microsoft started developing clustering for NT
servers.
Unfortunately, early implementations were
weak, even flaky. By Windows 2000 clustering was workable if not scaleable. With
Windows Server 2003 and Exchange 2003 all the pieces are in place,
clustering is now a feasible solution for fault tolerant mail servers.
There are two types of clustering. Active / Passive and Active / Active clustering. Always favour the Active / Passive or 'hot spare' solution. My reasoning is that you want to be sure
that second server can take the full
load. Where both servers are Active, at the very
least there will be a performance reduction, at worst the one remaining server will not be
capable of carrying the load of both servers. The Active / Passive is
like having a standby engine, whereas the Active / Active is like cancelling a
train and piling two loads of commuters onto one train.
The key development in Windows 2003 clustering has been to increase the
nodes from 2 to 8. The idea is to always have a 'hot spare' machine to
takeover, or 'failover' if any of the other 7 machines is unavailable.
The passive machine would have Exchange 2003 installed and would be ready to
seamlessly replace any of the regular Exchange Servers should they crash.
Guy Recommends: SolarWinds Engineer's Toolset v10
The Engineer's Toolset v10 provides a
comprehensive console of utilities for troubleshooting computer problems. Guy says
it helps me monitor what's occurring on the network, and the tools
teaches me more about how the system literally operates.
There are so many good gadgets, it's like having free rein of a
sweetshop. Thankfully the utilities are displayed logically: monitoring, discovery, diagnostic, and Cisco tools.
Download your copy of the Engineer's Toolset v 10
The key to configuring clustering is to focus on the Exchange services and the concept of Virtual servers. Clustering relies on shared storage. This shared disk takes the form of SCSI or SAN.
Let us say that the active node on server A is in charge of Exchange Services, for example System Attendant. If that active node fails then the passive node B (hot spare) takes over those services
previously hosted by server A. As the disk containing the mail is shared between all members of the cluster there is no transfer of data, and little latency in the Exchange services failing over.
Meanwhile the clients are still connecting to the virtual server name and IP. So the email users do not notice that the services have been switched automatically from server at node A to the server
at node B.
This section is only intended as an overview on configuring clustering.
My best recommendation on server hardware is, pick the brains of your
suppliers. They will have all the features at their finger-tips and
guide you to an integrated system with components designed to work in
clusters.
Those who use the HCL (Hardware compatibility list) as their Bible will
be rewarded with compatible components. In particular, be sure that
your RAID disks are cluster aware. Those who mix and match components
could be in for a nasty shock when they get messages like STOP 0x000000B8
due to hardware which is incompatible with clustering. Talking of
hardware in general and disks in particular, take the opportunity to
research other technologies like
SAN (Storage Area Network).
If you have to make the most of existing hardware, then the Windows
Server 2003 Help and Support will guide you through preparing your server.
You will need a second network card so that the cluster machines have their
own private 'heartbeat' network.
Incidentally, unlike Windows 2000, the clustering service is now installed
automatically in Sever 2003, there is no need to Add or Remove Programs,
just run CluAdmin. The wizard will then guide you through the creating
and naming of your cluster. What you will need is a service called MS DTC (Data Transaction Co-ordinator)
Always install and configure the cluster service before you install Exchange.
Sounds simple, but trust me if you do it the other way around it wrecks your
Exchange server. (Another reminder to always backup before you attempt
something new or adventurous.)
From the client's point of view
they see the cluster as a virtual entity so you must assign a unique name to
the cluster. The actual names of the servers are hidden from the
clients.
Guy Recommends:
The SolarWinds Exchange Monitor
Here is a
free tool to monitor your Exchange Server. Download and
install the utility, then inspect your mail queues, monitor the Exchange
server's memory, confirm there is enough disk space and check the CPU
utilization. This is the real deal - there is no catch. SolarWinds
provides this fully-functioning product for free, as part of their commitment to
supporting the network management community.
Clustering will be the way of the future. If
you are serious about eliminating single points of failure then research
Windows Server 2003's clustering. Always favour Active / Passive clustering and avoid Active / Active pairing if possible.
Recommendations
Choose Active / Passive clustering. Pay
particular attention to hardware compatibility before you buy your servers
with