6.2 Exchange Workload Simulation 451
Chapter 6
will receive some results to partners and employees who will participate in
the benchmark.
6.2 Exchange Workload Simulation
There are two families of tools to be used with Exchange 2003. One,
directly inherited from Exchange 5.5, is Microsoft Loadsim. Loadsim aims
at testing and generating load for MAPI clients only (primarily Outlook, if
you consider that applications that use the MAPI API can be considered cli-
ents). Because Exchange 2000 came with a significant emphasis on Internet
protocols, Microsoft engineered ESP (code named Medusa), a beefed-up
version of InetLoad. It is used to generate load and measure response time
for Internet protocols (e.g., POP, IMAP, and HTTP). ESP can be used with
Exchange 2003 and is particularly useful if you wish to stress SMTP mes-
sage routing, if your deployment has an important Internet protocol flavor,
such as for service providers, or if you deploy ActiveSync mobile clients.
For individual directory load simulation, you can use Loadsim, which
will simulate directory lookups via MAPI (it will also exercise the
DS_ACCESS component of the Exchange 2003 server), or ADTest, a spe-
cific tool that will generate and measure response time for concurrent
LDAP calls to an Active Directory server (GC server or domain controller).
6.2.1 Loadsim 2003
Loadsim was developed early along with Exchange 4.0; it provides a solid
tool and a de facto industry standard for simulating MAPI connections on
an Exchange 2003 server. Loadsim must be carefully used because it can
produce misleading results. For example, if your load simulation clients are
overloaded (!), they wont simulate as many users as you might assume.
Along time, and with experience building up, Microsoft improved both the
tool and the documentation and practices that came with it. Always make
sure you have the most recent version of Loadsim and the accompanying set
of documents that describe the mode of operations of the simulation tool.
Loadsim is particularly indicated if you wish to simulate as close to real-
ity a workload as possible. There are also certain testing scenarios (e.g.,
using mailboxes with a relatively large number of items in the critical fold-
ers of a mailbox) that can produce quite interesting results. We will discuss
the possible scenarios later in this section. You can download Loadsim from
the Microsoft Exchange Web tools page, available from: http://
452 6.2 Exchange Workload Simulation
www.microsoft.com/technet/prodtechnol/exchange/downloads/2003/
tools.mspx.
Key Components
You cannot just run Loadsim setup on your Exchange server and start kick-
ing simulation on the server. You have to organize a minimum environment
composed of the following items:
System under test (SUT): This is the system you wish to apply the load
on and possibly measure the response time of (known as the score) as
well as its behavior under activity. It is the server that runs Microsoft
Exchange and that hosts the mailboxes. You may configure Loadsim
to simulate clients to more than just one SUT. This is an interesting
approach because it also enables you to test the interserver communi-
cation.
Load clients: Typically, you will use between 500 and 600 MAPI con-
nections per loading client. Thus, if you wish to simulate 5,000 users,
you will need at least 10 loading clients. You must install Loadsim on
these clients, which do not need to run a client version of the Win-
dows operating system (e.g., Windows XP), they can also run Win-
dows Server 2003. You should not install Loadsim on the Exchange
server.
Monitor client: This is a special instance of Loadsim, which typically
simulates between 100 and 200 MAPI sessions and is used to record
the score.
It is common to configure Loadsim clients and measure the response
time across all of the load simulator. Also, you may use server-class
machines to run Loadsim and have a slightly better ratio of simulated cli-
ents than with a standard desktop (for example, 800 MAPI clients). How-
ever, we have found that the most predictable results were generated using a
lower number of simulated clients.
When Loadsim runs, it records the response time for the various MAPI
operations done (in milliseconds) and computes a weighted average
between all those response times. The score, final indicator of the SUT
response time, is generally the 95th percentile of the weighted average.

Get Microsoft® Exchange Server 2003 Scalability with SP1 and SP2 now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.