Optimizing Exchange 2003
5.1 Introduction to Exchange 2003 Optimization
As you deploy Exchange 2003 in an enterprise, you have to make consider-
ations for the consolidation opportunities and the necessity to place servers
in close proximity to the end users. With the introduction of cached mode
in Outlook 2003, the ability to centralize Microsoft Exchange servers has
been signiﬁcantly increased, and we now see deployments for large multina-
tional companies that are conﬁned to one or two data centers.
In the case where cached mode is not appropriate, online mode through
either Outlook or by the means of Outlook Web Access will require rela-
tively good network connectivity, providing both sufﬁcient bandwidth and
relatively low latency.
In either of these two deployment approaches, you must work on how
to best optimize server resource utilization and the Microsoft Exchange
software conﬁguration, in addition to server components such as CPU,
memory, and I/O (both storage and network related). The optimization
strategy, however, varies depending upon the type of deployment—whether
you adopt a scale-up solution that aims at gathering as high of a workload
on as small number of servers or decide to distribute the workload on a
number of servers. This latter approach is referred to as scale-out and calls
for the ability of the software components of Microsoft Exchange to prop-
erly distribute their operations.
Each approach is valid and may be mixed, depending upon the server
role. For instance, mailbox servers tend to scale up and host many users per
Windows server, while bridgehead servers can be easily scaled out, utilizing
less powerful machines (two processors, 2-GB RAM) but in several
instances, for helping on availability and easily scaling up if demand rises.
330 5.1 Introduction to Exchange 2003 Optimization
Each of these deployments requires a different type of hardware and
software conﬁguration, which is discussed in this chapter.
Some of the optimizations must be done regardless of the client type,
server role, and topology but more inline with the actual server hardware
For example, the use of clustering and the need for fast and reliable
failover requires necessary certain modiﬁcations in the Exchange 2003
STORE conﬁguration that are not present by default.
Optimization and tuning in Microsoft Exchange is typically done either
by means of a Windows Active Directory parameter change in the conﬁgura-
tion naming context and Exchange service container or using the server reg-
istry. There is a clear advantage in using Windows Active Directory–based
conﬁguration settings, as they can be done centrally, replicated throughout
the enterprise network, and picked up by the server next time the Microsoft
Exchange services are restarted. Some of these settings include most of the
optimization required for storage groups and database engines.
The optimization technique will not only depend upon the server role,
as we explained earlier, but also upon the mode of utilization of Microsoft
Exchange. If you use X.400 transport for connecting into legacy Exchange
5.5 servers or foreign mail systems, you will require some setting optimiza-
tion, which would not apply if you were to use native SMTP communica-
tion across the network.
Although this distinction was quite necessary as Exchange 2000 was
introduced, you will ﬁnd that most deployments are now using the default
SMTP transport, and we should be focusing on this transport for optimiza-
tion: in that case, the communication layer between the Internet informa-
tion service process and the STORE process should be as optimized as
possible, as well as any data processing that might be required, such as con-
tent conversion, SPAM ﬁghting, and virus detection/prevention.
Finally, some deployments require very speciﬁc optimization, such as in
service providers’ environments, in which a large concentration of users is
gathered, with small activity during most of the day, except for peak access
hours. The utilization of POP or IMAP clients can be improved with
Exchange 2003, saving precious CPU cycles used for handling connections
during peak periods. Another example is the optimization of the Windows
Active Directory topology such that minimal replication latency is found
between the servers used for provisioning and the Microsoft Exchange serv-
ers themselves. Backup and recovery tend to be fairly critical in these envi-
ronments, although we must accept the fact that messaging is now