Latency is the time it takes for a packet to travel across the network. Jitter is the variation in latency which you would think is zero but never is. Problems caused by latency and jitter are on the rise mainly due to the fact that they were never a consideration when protocols such as TCP were developed.
The rise in high speed WAN links means that people believe they have the bandwidth to share data remotely. The problem is that TCP does not always wait long enough for the packets to be received. TCP does not have different mechanisms for reporting a lost packet or a late packet, so both are resent and the receiver has to wait for this to be resolved. There are two major clues to these problems, one is the TCP retry rate and the other is the size of window TCP is trying to transmit. To look at both these issues, packet analyser technology (such as Network Instruments Observer) is required to see the detail. The better tools have enhanced organization of the data and can help with the diagnostics.
Once TCP protocols hit the limit of the latency they can cope with, performance drops off rapidly. The bandwidth tools (such as the FlukeNetworks RFC 2544 device) will not necessarily help here as they use UDP not TCP for testing and hence can only confirm that the contracted bandwidth is available.
The bandwidth calculation tools (such as AppNeta PathView) however are more useful as they are aimed at measuring TCP performance and will give a more realistic real time throughput.
Jitter is a measure of how consistent the packet delivery is to the end point. Ideally once the delays in the network have been traversed, this will be a consistent number, but it rarely is. The applications which are heavily affected by Jitter are the Voice and Video applications as these require consistent delivery of data to provide a smooth picture or a clear voice.
A certain amount of jitter can be smoothed out (known as a Jitter buffer) but if exceeded the frames are dropped. As these technologies do not have time to replace the missing frames become gaps in the voice or picture.
There are specific measurements to analyse the performance of such applications such as R-Factors, MoS scores etc. Some products (such as Fluke Networks OptiView and Observer) have specific tools for Voice analysis and others (such as AppNeta PathView) can measure the networks ability (both pre- and post-deployment) to perform whilst running such applications.
Return to Article Go on to Layer 3 Design
Last Updated: 14/06/2011