The rise in Wireless and DSL technologies has lead to an increase in dropped packets and many modern applications don’t respond very well in this environment.
Wireless has its own layer 2 correction technologies hence there is a great deal going on under the surface to try and make sure the packets get through. Monitoring these specific errors is very helpful when looking at performance problems as they tell you how hard the technology is having to work to get the data frames delivered.
Standard SNMP tools do give some of this information, however a whole generation of specific tools has emerged (such as AirMagnet & Fluke Networks EtherScope) to do this in more detail. Wireless also brings it own questions regarding coverage and connection issues which again require specialist tools (such as AirMagnet) to diagnose.
Once the layer 2 technologies have got the packets through, the TCP technologies take over, and issues can be much greater due to the delays associated with getting the packets this far. Troubleshooting at this layer again falls to the packet analyser technologies mentioned in the Errors section above.
In looking at the end to end performance of any link people start by looking at the stated bandwidth of the technologies involved and expect that they are going to get something just less than what is says on the tin. However, there are ways of measuring the actual performance of a path so you can understand what is genuinely achievable and (especially in the case of Wireless, DSL and any shared media) how stable the top end performance is.
There is a standard for measuring link performance called RFC 2544 and this provides a method of measuring the actual bandwidth available (such as Fluke Networks EtherScope). This is a standards based test, hence very powerful when in discussion with the carrier and reviewing the provided capacity. It must be noted that this technology needs the link out of service and can take 20 minutes or more to complete, so is not practical during the day.
There are emerging technologies (such as PathView from Apparent) to consider that can calculate the performance of a live link using one ended tests, simply point at any IP address to discover the realistic performance achievable to that end point, including detail on each hop of the route.
Return to Article Go on to Latency
Last Updated: 17/08/2010