Forum

The cheapest quick way to measure throughput is iperf.

http://dast.nlanr.net/Projects/Iperf/
Look down the page for user contributed, compiled, packages.

You could put it on a machine on the same subnet as the suspected problem server to see if throughput to that subnet is different from others.

By running iperf at different times of the day, you could see if there is a network load problem.

Using Ping with long packets continuously for a minute or two can show dropped packets and high variability in response time, indicating router or other network congestion.

Visualroute has clients and servers that let you test throughput and track TCP delays as well as jitter in some of their software.

http://www.visualroute.com/

Once you isolate TCP "conversations" in protocol analyzers like Wireshark or OmniPeek, you can look at timing for client-server responses. OmniPeek has built in timing thresholds for slow application servers that I've used successfully to show that there was an application server slow response vs WAN problem. Recently I had some problems measuring throughput using Wireshark on Windows, and I got more believable information with OmniPeek. It could have just been my problem, and I've used earlier Wireshark (Ethereal) successfully on Linux.

Charles Preston
Comsec Wireless