How Does Packet Loss and Delay Affect Performance?
Tony Fortunato of The Technology Firm walks you through the process of evaluating the impact of packet loss and delay on a network's or app's performance.
April 28, 2022
When troubleshooting performance issues with a client the other day, I was explaining that packet loss or delay will affect your performance. The question is, how much loss or delay are you willing to tolerate, or how much of a performance hit are you willing to take before you investigate?
Since specific packet loss and delay are hard to recreate, I used my Linktrophy Mini wan emulator from Apposite Technologies to help me out. Any wan emulator where you can change the loss and delay will do.
An alternative to consider is to modify your Packet Loss and Latency within VMWARE Workstation, under Network Adapter Advanced Settings
I explained to the client throughput is a fairly general term, and the first thing I need to do is define what I'm measuring to avoid confusion. For example, using iperf to measure network performance is totally different than copying a file to your network drive using SMB. In the 'real world,' you have all sorts of items that impact performance. For example, distance/equipment/processing latency, cabling errors, and interference are just a few examples. Then there are the techniques we use to mitigate performance loss like full-duplex communication, multiple flows, hardware-level encryption, non-mechanical drives, etc.
In this example, I used iperf to measure the impact on network throughput while introducing latency and loss.
The point of this is to make you aware of what can impact your performance as well as tools and methodologies you can consider when you test stuff.
The results are fairly self-explanatory, as you can see from the table below and as I explain in the video.
What was interesting in the results that I could present to the client was the combination of loss and the direction of the traffic. His issues were more delay-based, but the table still put the point across.
I tried to use conservative numbers, but I always encourage you to figure out your specific values to get more accurate findings.
Related articles:
About the Author
You May Also Like