Windows 2003 Terminal Server

[ionoxx]

Alright, I have 2 Windows 2003 Servers running terminal services for 40 client computers. Every once in a while they slow down. We reboot the whole system including the domain controller and it usually fixes the problems. The CPU time and memory usabe is no more that 10-15% at anytime. The network is not a bottleneck either. It usually happens that only one of the two servers is slow. I've been looking through performance log data all morning and found one thing.

Terminal Services Session\Output WaitForOutBuf

Described as "Shows the number of times that a wait for an available send buffer was done by the protocol on the server side of the connection."

Some of the connections objects show very high numbers, in the thousands yet most connections only have 45 to 50.

Thanks for the help.

 

[Woodie]

You're rebooting three servers, including a Domain Controller, just because some users are getting slow response???

OK...start off small. Reboot or perfmon the box that they're having trouble with. Reboot one box at a time, until you can identify the server that's having/causing the problem. Especially when you say that only ONE of the two TS servers is having a problem.

Based on your perfmon info...sounds like a few clients have poor connections, so are causing the Output WaitForOutBuf counts to climb. Can you map back the problem (high-count) connections to clients or users?

 

[ionoxx]

Well, we reboot the servers cause its really slow. Otherwise we wouldn't bother. Performance monitor hasn't returned anything useful thus far. Connection problems is doubtful, cables have been there for years without problems. Latent collisions are not a problems, collisions are not a problem either. The problems are intermitent in any case.

 

[Woodie]

How do you know the connections aren't a problem? Is it an internal network?
Cables being there for years is a two-edged statement! LOL

What stats are you basing the statement that "memory is not a problem" on? (Taskman, perfmon, etc..)

Any hits in Technet on that text?

Is there a particular thresh-hold in terms of number of TS sessions when you get spikes?
Any correlation between the users saying they are getting slow response, and the spikes in perfmon for the "send buffer"?

 

[ionoxx]

We know the cables are good because the problem is not from the data transfer, its not the response of the GUI, its the response of the applications running on the server ( applications like Office and Agency Application software).

It's an internal network, with 40 clients, staticly assigned to connect to one of the two servers. So no more than 20 clients on each server. And the clients are connected all day, and log off at night, so the number doesn't change as the day goes along.

As for the specs of the systems, they are both Dell PowerEdge 1600SC servers with Dual Intel Xeon 2.4GHz processors (HT enabled), 4Gb Ram, 6Gb page files and 18Gb SCSI drives in raid 1. Performance monitor reports no more than 10-15% usage of the processor and no more than 1.5Gb of used memory at any time.

The domain controller which also runs the database for the application software being used, is a Dell PowerEdge 2500 with an Intel Pentium III 1.0GHz, 1Gb ram, and 3x 18Gb 15000rpm SCSI drives in raid 5. This server uses no more than 50% of its processor usage and no more than 300Mb of ram at any time.

TechNet returns the description that i mentioned above and that is it. A google search returned nothing useful either. Thats why i'm here.

As for the correlation, i'm working on it.

I hope this helps! Thanks for the help so far.

 

[djdrastic]

Just a weird question ?

How are the printers mapped on the TS ?

 

[ionoxx]

There are 5 network printers, TCP/IP connections to them. And there are 3 printers that are being shared from the workstations that are running Windows 98. Those are the same workstations that connect to the TS.

 

[djdrastic]

Are the printers visible in a TS Session ? Aka are they added as local printers on the server or are they just brought up in a session ?

 

[ionoxx]

They are local printers on the servers.

 

[casper114]

It doesn't fix the problem when the Primary Domain controller isn't rebooted? Or when all the machines are rebooted except for any one machine?

 

[dphantom]

What is "once ina awhile"? Do you see a pattern where after x number of hours, the response slows. Is it always in the afternoon, mornings, etc...



Are all clients slow, just the clients on one TS slow or a mix of clients? If you ran the app or apps from teh console of the TS, do you get a good response when others report slow?

Why in the world would you reboot a DC and why is a DC hosting a database anyway???

What kind of switch/hub infrastructure do you have? Switches I hope.

 

[dphantom]

After further research, maybe this is what is happening.

Output WaitForOutBuf Shows the number of times that a wait for an available send buffer was done by the protocol on the server side of the connection.

The above is what the performance indicator says.

The architecture of your system is this as I understand it.

The server is your DC running a database.

The client is your TS servers

So the DC is attempting to send data but is getting so many requests that its send buffers are filling up because the TS servers can't send data fast enough to the end points.

So a couple more questions.

What type of database is running your application? If Access, then I can see lots of problems with 40 users. If SQL/Oracle/Sybase, is the databse designed to handle that many users?