That’s the question that I was asked today.

My answer: It depends.

Many folks will ask if we can share the results of tests that we’ve performed with other customers.  I’m sorry, but we can’t.  The vast majority of the testing that we do is specific to that customer (many times, applications that they have written), so results vary.  And they should.  Different customers utilize our OS in different ways, so testing what each customer is doing is a very specific thing.  Yes, we do run into a lot of the same issues and implement some of the same solutions, but the results of one customer will be different than another.

Also, when speaking with any customer about performance, I am always careful with making general statements about performance (networking or otherwise) as with any customer there are many, many variables that should be taken into consideration, including, but not limited to:

  • What type of traffic
  • Characteristics of that traffic
  • Number of *concurrent* connections

Begin Sidebar
Many customers don't truly know the correct definition of a "concurrent" user. When it comes to the way we perform testing, two concurrent users are “users that are actively transferring data at the same time”. If you have a server that has two users connected (via viewing a web page for example), they are not concurrent unless they are actively transferring data at the same time. Even though a customer may have 5000 users, how many of those users are *concurrently* using the the system will usually be a lot smaller. However, most customers when discussing network connectivity and performance usually don't think concurrency in this manner. Instead, they think of concurrency as the total number of simultaneous TCP connections.
End Sidebar

You may want to take a look at this webcast which does a good job explaining the different areas of Windows 2003 Server performance improvements. The speaker does a great job of discussing the different areas such as networking, windows media server, web, sql, etc.

http://www.microsoft.com/seminar/shared/asp/view.asp?url=/seminar/en/20030424vcon02/manifest.xml

As the speaker states, "On x86 we can support around 100,000 simultaneous tcp connections on any given system." He further mentions that with ia64, we can support over 1 million connections.

Also, here's a handy tool if you need a little help with some of your performance related tasks.  It does a fairly good job on making suggestions on how to improve performance:

http://www.microsoft.com/downloads/details.aspx?FamilyID=09115420-8c9d-46b9-a9a5-9bffcd237da2&DisplayLang=en

Microsoft ® Windows Server ™ 2003 Performance Advisor is the latest version of Server Performance Advisor, which is a simple but robust tool that helps you diagnose the root causes of performance problems in a Microsoft Windows Server 2003 deployment. Server Performance Advisor collects performance data and generates comprehensive diagnostic reports that give you the data to easily analyze problems and develop corrective actions

Microsoft ® Windows Server ™ 2003 Performance Advisor provides several specialized reports, including a System Overview (focusing on CPU usage, Memory usage, busy files, busy TCP clients, top CPU consumers) and reports for server roles such as Active Directory, Internet Information System (IIS), DNS, Terminal Services, SQL, print spooler, and others.
 
I would highly suggest to anyone concerned about performance to perform your own testing (with our assistance or not) to determine the levels at which your systems will handle the particular tasks that you will be performing.
 
-Marcus

Contact your TAM, ADC, MCS Consultant or other Microsoft Representative to find out how you can use our facilities