Tip of the Week: Measuring Network Latency

Blog Post created by ken_foster on Feb 23, 2014

Network latency is of great concern when using Mentor Graphics software, particularly the releases that have an underlying iCDB architecture. The ping utility is often used to get an idea of how the network is performing. However, the ping results are sometimes not granular enough, and the data can be difficult to analyze if you run ping for long periods of time.


In this tip of the week, I'll tell you about one of my favorite tools for getting good network latency data: psping.


To get useful network latency measurements you should use a larger packet size that closely represents the packets created by the application you are verifying, and you need to do it over time. You can do that with the standard ping command, but then you'd have to post process the data to get the information you want. Doable, but not desirable. One thing that you can't do with the standard ping that is sometimes useful is ask it to use a specific port number so you can verify that the port you want to communicate with is open. Psping does all that, and provides a tidy summary report.


Psping works in client/server mode for the network latency test. In this example, I'll start a psping server on a machine called 'kvmw7x64', and ask it to listen for pings on port 9001 using TCP protocol. I've also specifed IPv4:




Now on a client machine, I'll start psping and ask it to transmit 1200 byte packets to kvmw7x64 on port 9001 for 3600 seconds (1 hour):




When the hour is up, I can view the results as a summary. This provides a pretty good indicator of what my actual network latency is over time.



I'd suggest you fire this up during the time of the day when your network is the most heavily utilized and again when it's the least heavily utilized. Save these numbers as a baseline. It might not hurt to run it every so often to see if things are changing. Then, any time you suspect problems you are having are related to a slow network, run it and compare the results to your baseline. That should either incriminate your network exonerate it.