Question

I have .Net service that listens on single port over TCP protocol. Clients connect and then transmit data for some time (from few minutes to several hours).

Is there any limit on number of connections on Windows 2008 server? I did not hit any, since now there is up to 50 users. Plan is to have thousands of users, so I’d like to know if there will be problems in future.

Edit:
As Cloud answered, it seems that there are some limits in some versions of Windows Server 2008. Is there any reference on those limits? I tried Google, but it returns articles on limit on half-bound tcp connections.

Answer 1

How many thousands of users?

I’ve run some TCP/IP client/server connection tests in the past on Windows 2003 Server and managed more than 70,000 connections on a reasonably low spec VM. (see here for details: http://www.lenholgate.com/blog/2005/10/the-64000-connection-question.html). I would be extremely surprised if Windows 2008 Server is limited to less than 2003 Server and, IMHO, the posting that Cloud links to is too vague to be much use. This kind of question comes up a lot, I blogged about why I don’t really think that it’s something that you should actually worry about here: http://www.serverframework.com/asynchronousevents/2010/12/one-million-tcp-connections.html.

Personally I’d test it and see. Even if there is no inherent limit in the Windows 2008 Server version that you intend to use there will still be practical limits based on memory, processor speed and server design.

If you want to run some ‘generic’ tests you can use my multi-client connection test and the associated echo server. Detailed here: http://www.lenholgate.com/blog/2005/11/windows-tcpip-server-performance.html and here: http://www.lenholgate.com/blog/2005/11/simple-echo-servers.html. These are what I used to run my own tests for my server framework and these are what allowed me to create 70,000 active connections on a Windows 2003 Server VM with 760MB of memory.

Edited to add details from the comment below…

If you’re already thinking of multiple servers I’d take the following approach.

1. Use the free tools that I link
   to and prove to yourself that you
   can create a reasonable number of
   connections onto your target OS
   (beware of the Windows limits on
   dynamic ports which may cause your
   client connections to fail, search
   for MAX_USER_PORT).

2. during development regularly test your actual server with test
   clients that can create connections
   and actually ‘do something’ on the
   server. This will help to prevent
   you building the server in ways that
   restrict its scalability. See here: http://www.serverframework.com/asynchronousevents/2010/10/how-to-support-10000-or-more-concurrent-tcp-connections-part-2-perf-tests-from-day-0.html