Jitter update: Measuring Network Quality
This is an update to a previous jitter posting. The last posting was about the FUD factor in network performance solutions like QOS. This is a more organized look at the topic of network performance. Network performance can be evaluated by measuring how much, how fast, and how accurately the network traffic flows through the network. There are various types of traffic such as video, data and voice. Each type of traffic could have a very different quality measure on the same network. There are also two types of measuring that can be done—active and passive. Active monitoring puts traffic on the network at regular intervals and measures quality. Passive monitoring polls various network devices and collects information that they accumulate about how they are performing.
Currently UEN applications include things like distance learning via two way video, internet access in public libraries, a variety of web based services such as pioneer library http://pioneer.uen.org/, PowerSchool http://www.apple.com/education/powerschool/, SURWEB http://www.surweb.org/, research in a variety of areas, and increasingly telephone calls over IP. UEN does active measurement of circuit failure, and passive measurement of a variety of router variables including usage, router CPU, and router reloads. What would be interesting to measure is what our users see such as video and audio quality, web response time, and file download speed.
Sometime I think that network monitoring is like the statistician and the light pole. One evening a group of people were walking along and one of the party dropped an ear ring. The party commenced to look for the ear ring. All but one of the party began looking in the area that the item was dropped. Finally somebody asked the statistician why he was looking for the ear ring some distance away under a street lamp. He replied that the light was better there. I think we do passive measurements of router variables because that is what our network monitoring packages do.
Doing a Google search on the subject I came up a company, Brix Networks–http://www.brixnet.com that does both active and passive network monitoring. In addition, the active measurements are protocol specific. You could, for example send an H.323 packet between nodes and find out the latency each way, jitter, early packets, and late packets. If that we all it did, it would be an interesting tool. In addition however, it runs a score which is highly correlated with the Mean Opinion Score (MOS score). MOS is calculated by having a bunch of people see video and listened to audio and rate the quality on a scale of 1 to 5. “Toll quality” is given a score of 4. Making the measure of network quality the quality of the networks user’s experience is, I believe what network performance is all about.
The same equipment that can be used to monitor the network can also be used to test specific configurations. You can on a test network have the Brix equipment send a traffic load of some size with a traffic mix that matches your network, so much H.323, SIP, web traffic, email, or whatever and measure the various network quality scores. Then you can see what would happen if your H.323 traffic were to triple. You could measure the effects not only on the video, but on all of the rest of the traffic. You could measure the effects of different queuing and prioritization strategies like FIFO, each flow get the same priority (called weighted fair queuing by Cisco), video and audio traffic get first priority or any other scheme that you want to test.
There are many theories about how to achieve a quality network. Here are the two suggestions that pop to the top of the stack.
- Make sure that there is 70% headroom in the circuit.
- Replace all networks with routers and switches that are QOS enabled so that you can set and enforce priorities that will be used to make some traffic go better and some traffic go worse.
Both suggestions have costs associated with them. Both are difficult implement and enforce. Just one attack like the slammer can fill up your headroom. Perhaps your network budget gets cut and you can only afford 10% headroom. Just one hole in your trust boundary can let packets be classified by an intruder. Maybe you don’t have capital dollars to buy all new routers and switches. These suggestions are made in an environment where there is no way to see the effects on the quality except to have each person see what they get on their favorite application. When user complaints come in you are left with guessing and blaming.
I have heard of several other companies beside Brix Networks that have similar products. I find the strategy of monitoring something very close to what your users see to be intriguing.
3:54:11 PM 
|
