how to detect NT network slowdown causes

You can use the performance monitor that comes with NT server, if you can, use the one that comes with SMS cd.

If you have funds try HP openview, and Cabletron has a program called spectrum.

Make certain that browser wars are not occuring..., With that size network there should only be one Master Browser, the server. To stop any elections, stop the service on all workstations...,

What I would do first of all is to check the workstations are all at 10MB and not at 100MB, and that the server is at 100MB full duplex.If not, try those settings, it should help!

Does this apply to all workstations or just some? And when does it happen? Is there any pattern to the problem?

If it does boil down to network, I would check the switch first. This, because of delays in logging on. Confirm that ports on the switch are properly assigned.
But, in order to be on the safe side about the network - try copying a large file between the server and a workstation and time it. After that, you would have a clearer picture - network or not.

First off, if the workstations are Dell's also, you may be running into a problem we had at our company - The brand spanking new Dell OptiPlex GX1s that came in would exhibit similar behaviour until we found out that there was an issue with the 10/100 card installed in these workstations. Apparently, every so often they would get confused and try and renegotiate the speed of the link between themselves and the switch, resulting in miserable performance for up to a minute. Even if you don't have Dell's, I will give you the same suggestion, which is that you should turn off autonegotiation (since you have a switch, which will do store and forward for you, as well as the fact that you have all 100bt capable cards - negotiation is moot). You should configure each adapter to specifically be 100bt/full-duplex (if your switch can handle full-duplex - may as well take advantage if it can), and if you have a manageable switch, configure each port to be explicitly 100bt/full-duplex. This should eliminate any pr

On the surface it sounds like the hardware should be up to the task. I had a similiar problem when implementing Outlook 98. It could be a name resolution issue. I have seen similiar problems when running DNS and WINS. Outlook will use WINS for login and communications, and failover to DNS if WINS has a problem. This can cause a significant delay in response times.

If you use an NT domin name that matches the DNS domain name, check to see that the DNS "Domain" name in the NT Workstation Network control pannel is a fully qualified domain name ("company.com" not "company").

Try to ping the server name, is there a delay in getting the ip address? If so this could indicate that WINS is handing off to DNS. You could delete the WINS database and have it rebuild a new one. Stop WINS. Delete everything in the WINNT/SYSTEM32/WINS directory. Restart WINS. THis will cause a new database to be built as the server and workstations make inquiries.

Refer Microsoft article Q185786 forrecommended WINS pra

Look for the most likely problems first, and those would be either answer #5 or else a bogged down server. You can verify the latter by running some perfmon tests on it. Althought it sounds like your workstations are robust enough, the Dell PowerEdges with SBS have Exchange loaded on them, and trust me, Exchange is a resource hog! With Exchange services running, you will want as much memory as possible with a fast hard disk (you do have SCSI, right?). Look for peaks in usage, i.e.--does everything get slow when a large print job is submitted? Does it happen at 8:00 AM when your users are logging in? Do you use large login scripts? Windows NT mag has good suggestions for acceptable perfmon counters at www.winntmag.com

Good luck.