SUMMARY
This step-by-step article describes the implementation of
Exchange on a Microsoft Windows 2000-based cluster.
Although other
information is available about Windows 2000-based clusters, this article
describes server clustering from the Exchange viewpoint.
back to the top
Exchange Clustering Overview
When you install Microsoft Exchange 2000 Server on a Windows
cluster, it is configured as a virtual server. It runs as a virtual computer
that can be passed between cluster nodes. Exchange clients are unaware of this,
and only have to connect their Microsoft Outlook profiles to the name of the
Exchange virtual server.
Clustering technology makes it possible for
you to design high-availability solutions for enterprise-level programs. This
is especially desirable for mailbox and public folder servers. However,
clusters that are based on the Windows 2000 Cluster service are different from
load-balancing clusters. Load-balancing clusters are suitable for Exchange
front-end (FE) servers that do not store any user data. Clusters that are based
on the Windows 2000 Cluster service rely on completely different mechanisms.
They are suitable for Exchange back-end (BE) servers that host the user data.
The Cluster service is available only in Windows 2000 Advanced Server and
Windows 2000 Datacenter Server. Note that only the Enterprise Edition of
Exchange 2000 Server supports this form of clustering.
For additional
information about the Microsoft Cluster Service (MSCS), see the "REFERENCES"
section of this article.
back to the top
Install Exchange on a Cluster
When you install Exchange on a cluster node, the Exchange Setup
program copies the program files to the local disk of the cluster node. If you
have a cluster with two nodes such as Node A and Node B, Setup copies the
Exchange program files to the hard disk of Node A. For example, this might be
drive C of Node A. Setup makes no changes to Active Directory at this time. You
then run Setup a second time to install the Exchange program files on the local
hard disk of the second node. For example, this might be drive C of Node
B.
After the Exchange program files are copied to the hard disks of
each node, you must configure a "Group" for the Exchange resources. A "Group"
in this case is defined as a logical container that holds resources for a
virtual server. This Exchange virtual server will have many of the same
resources that all physical Exchange servers have such as:
- A host name
- An IP address
- Disk storage
After you create these resources, you must configure resources
that reflect the Exchange services. For example, an Exchange System Attendant
resource, an Exchange Information Store resource, and other resources. These
resources can be either taken offline, which mimics the stopping of the
particular service, or brought online, which mimics the starting of the
particular service. You can also change the current resource owner (the cluster
node). For example, you can configure Node B to run this resource instead of
Node A.
The Microsoft System Attendant resource is the only resource
that you must manually create. The other Exchange service-related resources are
automatically created after you create the System Attendant resource. It is at
this time that changes are written to Active Directory and an object for this
Exchange-based server is listed in the
Servers container for your administrative group.
Exchange
supports two different configurations of virtual servers, Active/Passive and
Active/Active. For additional information about the installation of Exchange,
see the "REFERENCES" section of this article.
back to the top
Active/Passive Clusters
Active/Passive clustering is supported by Exchange Server 5.5 and
later. In an Active/Passive cluster, the following configuration exists:
- At least two nodes. For example, Node A and Node
B.
- One shared disk system, the ownership of which can be moved
between cluster nodes.
- One IP address that is created in the Exchange
"group."
- One network name that is created in the Exchange
"group."
- One virtual Exchange server. Both nodes can own this
virtual Exchange server, but only one node can own the virtual Exchange server
at a time.
In an Active/Passive cluster, clients connect to an Exchange
virtual server that is currently owned by one node of the cluster. For example,
this might be Node A. In this case, Node A has control over the shared disk
system that contains the Exchange databases. If Node A experiences a hardware
failure and goes offline, the other node in the cluster (Node B) detects this
and takes ownership of the virtual Exchange server and all its associated
resources. These resources include the network name, IP address, the shared
disk system, the Exchange System Attendant, and other resources. This process
occurs without the currently-logged-on users knowledge. Because the clients
connect to a virtual Exchange server, they are unaware of the fact that the
physical server has changed. After the offline node (Node A) is restored to
service, the virtual Exchange server can be moved back to it.
back to the top
Active/Active Clusters
Active/Active clustering is supported by Exchange 2000 Server and
later. However, this form of clustering is not supported in earlier versions of
Exchange Server. In an Active/Active cluster, you have multiple virtual
Exchange Servers instead of only one virtual Exchange server on an
Active/Passive cluster. Both of the virtual Exchange servers can be moved
between the different nodes of the cluster, or both virtual Exchange servers
can simultaneously run on one node. In an Active/Active cluster, the following
configuration typically exists:
- At least two nodes. For example, Node A and Node
B.
- Two shared disk systems, the ownership of which can be
moved between the cluster nodes.
- Two IP address that are created in two separate Exchange
"groups."
- Two network names that are created in two separate Exchange
"groups."
- Two virtual Exchange servers. Both nodes can own these
virtual servers and both virtual Exchange servers can run on either of the
nodes, but only one node can own each virtual Exchange server at a particular
time.
In most cases, each virtual Exchange server is running on a
separate node. For example, one virtual Exchange server would run on Node A,
and one virtual Exchange server would run on Node B.
In the case of
a hardware failure or where one node is taken offline, the other node detects
this, and then takes ownership of the whole Exchange group that was running on
the first node. This is known as a "failover." In the case of a two-node
cluster such as Node A and Node B, the second node (Node B) would then host two
virtual Exchange servers. In this example, Node B would have ownership of both
shared disk systems, both IP addresses, both network names, and both virtual
Exchange servers. The connected Exchange clients are unaware of this
process.
When the first node (Node A) is brought back online, the
virtual Exchange server that "failed over" to the second node can be moved back
to its original node. This process is known as "failback."
NOTE: Exchange 2000 Server on Windows 2000 Datacenter Server is
supported in an Active/Active cluster only in a two-node configuration.
Active/Passive clustering on Windows 2000 Datacenter Server has the following
configuration:
- Active/Active/Active/Passive
In this configuration, one cluster node is passive and is
available to host an Exchange virtual server if a failover occurs. As of
October 2002, a cluster with four active nodes is not supported.
Exchange 2000 Enterprise Server also supports a maximum of four storage groups
with five databases for each storage group. This limitation also applies to a
clustered node. Because of this, you must make sure that if all nodes fail
except for one, the total number of storage groups on the remaining node does
not become more than 4, and that the total number of databases does not become
more than 20. For example, in a two-node Active/Active cluster, two virtual
Exchange servers each with three storage groups would experience a problem at
failover, because the remaining node would have to host 6 storage groups, and
this is over the limit.
back to the top
Active/Passive Compared to Active/Active Clusters
There are advantages and disadvantages to both Active/Passive and
Active/Active clustering. Microsoft recommends that you use Active/Passive
cluster configurations because they are highly scaleable and they have less
administrative overhead than Active/Active clusters.
The advantage of
an Active/Active over Active/Passive clustering is that both nodes in the
Active/Active cluster work at the same time. However, because an Active/Active
cluster can support only 1,900 MAPI clients per node, and because of the
increased administrative overhead you must perform to maintain the cluster,
this advantage may not occur in a real-world scenario. For example, increased
administrative overhead includes monitoring for virtual memory
fragmentation.
The Active/Passive cluster configuration has less
administrative overhead, is more scaleable than an Active/Active configuration,
and because the Active/Passive cluster can host more users, it also has a
financial advantage.
NOTE: While the hosting of public folders is supported in a clustered
environment, it may not make financial sense to do so because you can easily
replicate public folders between the Exchange servers in your organization.
However, because mailboxes cannot be replicated, it is important to host
Exchange servers on which mailboxes are stored in a clustered
environment.
back to the top
Exchange Components Supported in a Clustered Environment
Not all Exchange components work in a clustered environment. The
following lists describe the Exchange components that are supported in a
clustered environment:
On an Active/Passive Cluster
- The Microsoft Exchange System Attendant.
- The Microsoft Exchange Information Store.
- The following protocols:
- Full-text indexing.
- Message Transfer Agent (MTA).
On an Active/Active Cluster
- The Microsoft Exchange System Attendant.
- The Microsoft Exchange Information Store.
- The following protocols:
- Full-text indexing.
- Message Transfer Agent (MTA).
NOTE: The MTA resource is an Active/Passive cluster resource. There
can only be one MTA resource per cluster. In the case of an Active/Active
cluster, only one virtual Exchange server will have the MTA
resource.
Unsupported Components
The following Exchange resources are not supported in a clustered
environment:
- Network News Transfer Protocol (NNTP)
- Key Management Server (KMS)
- Instant Messaging (IM)
- Chat
- Connectors to the following messaging systems:
- Novell GroupWise
- Lotus Notes
- Lotus cc:Mail
- Microsoft Mail
- IBM OV/VM (PROFS)
- SNADS
- The Microsoft Exchange Event
- The Site Replication Service (SRS)
For additional information about Exchange components
that are supported in a clustered environment, click the following article
number to view the article in the Microsoft Knowledge Base:
259197
Status of Exchange 2000 Server and Exchange Server 2003 components on a server cluster
back to the top
Troubleshooting
When Exchange is hosted on a Cluster service computer, you must
perform some common Exchange tasks differently than if Exchange is installed in
a non-clustered environment.
- Exchange 2000 Server Setup
Before you install Exchange
2000 Server on a cluster server, see the C20_clustering.rtf document that is
stored in the DOCS folder on the Exchange 2000 Server CD-ROM. - The Eseutil utility
You must give special
consideration when you run the Eseutil database integrity utility with the /CC switch. For
additional information about this issue, click the following article number to
view the article in the Microsoft Knowledge Base: 266689
The 'ESEUTIL /CC' command does not work on cluster server
- Adding additional Exchange virtual servers For additional information about
this issue, click the following article number to view the article in the
Microsoft Knowledge Base:
293510
How to add an Exchange 2000 virtual server to a cluster server
- Installing hotfixes or service packs
Read the README
file that is included with the update before you install a hotfix or service
pack on a cluster server. In most cases, the passive node of the cluster is
updated first. The virtual servers are then moved to the passive node, and then
the other node is updated. - Exchange database restoration
For information about
disaster recovery for Exchange 2000, visit the following Microsoft Web site:
back to the top