Business Library

| Rules of Thumb | Analytical Techniques | Simulation | Benchmark |

Having worked out what your network is going to do and produced an outline design for it - how do you predict its performance?

The choice of technique will be based on the amount of time you have to assess the performance, and how costly an incorrect decision could be. The graph to the left illustrates the dilemma, and introduces the main network verification techniques:

No Analysis is the first option, in which case it costs nothing to perform but the risk of getting it wrong is high!

"Rules of Thumb" are guidelines gained from experience, and represent a first attempt at modelling the proposed network, typically using a simple spreadsheet. For example, if you were to calculate the average amount of traffic expected to cross a particular communications line, you may not worry provided the media would be utilised at less than a certain percentage. [E.g.- if you needed to transport an average traffic volume of 10,000 bits per second across a line rated at 64,000 bps, this would exhibit an average utilisation of 10/64=16%.] This technique helps towards a greater understanding of the network and a reducted risk.

Analytical Techniques are used to calculate the performance of the network mathematically, normally using a purpose-built modelling tool. You have seen a simple example of this in the article on LAN Performance. Note that, even here, quite a lot of information is needed about the network and its traffic in order to make a sensible calculation - although this does start to consider the interaction between multiple users, hence further reducing the level of risk.

Discrete Event Simulation models the behaviour of a network over a period of time - typically the "busy hour" (unlike analytical techniques which typically calculate performance at some idealised instant) by mimicking the transport of messages around the proposed network configuration. So if, for example, we take a communications line, a simulation will take into account waiting for the media to become free by holding the message in an output buffer. When the line becomes available, the time to transport the message is calculated, as is the effect of any overheads. Taking the total size of the message, the bandwidth of the link, plus any delay due to propagation (normally only significant over long distances), the simulator can calculate when the message will arrive at its destination and enter the input queue of another device. At this point the link is free and available once more. The time for an individual message to be sent and arrive at its destination can therefore be tracked, taking into account any other messages occupying resources necessary for it to proceed. In reality, the simulation would be for the entire network - or a significant portion of it - rather than just a single component as described here. Notice that this technique is much more complex than analytical techniques - and considerably more information is required about the network and the traffic characteristics. However, the accuracy can be very high.

Benchmark is the ultimate model of a network - the operation of the actual network configuration, or more typically a subset of it, in a laboratory environment. To be realistic, the benchmark environment will require sufficient hardware to exercise all the proposed network equipment and the ability to generate traffic which relates to the load the real network will ultimately be required to carry. This may include getting a number of end users in to exercise the system with real business transactions - a typical example might be a financial dealing room where the dealers operate a dummy environment using dummy applications which are as close to the real system as possible.

As each of the four techniques is more complex and requires more detailed information about the network - particularly the expected traffic - they cost more in terms of the equipment needed, the modelling tools required, and people. But what are you achieving? The green line in the chart illustrates the point: A more detailed model produced prior to installation will help reduce the risk of it failing to deliver the required levels of service.

"Which technique do you use when?" Remember that this is a question of time and cost versus the cost of getting it wrong. In a simple LAN environment, the cost of additional hubs, switches, and cables might be relatively small - hence only a basic analysis need be performed - rules of thumb or analytical techniques. WAN networks which interconnect many sites will be much more difficult to change significantly once in place. Also, if your network will carry critical data, such as air traffic control information, you will be considerably more concerned about any potential risks of the network failing to deliver!

You must also consider the accuracy of the input information, particularly with regard to the predicted traffic, as this is a classic case of "Rubbish In, Rubbish Out". The increasing accuracy of each technique depends on the increasing accuracy of input data. You might need to measure the traffic on an existing network in order to make the estimates more reliable - and how this is done is a subject in its own right.

There are several suppliers of specialist network analysis tools and techniques. For example, the CACI Web site illustrates the pertinent points in this article in terms of a pair of network modelling tools:

1. Comnet Predictor enables a mathematical calculation of network performance to be made quickly and easily.
2. Comnet III is a network simulator which can be used by itself, or in tandem with Comnet Predictor.

Both tools provide ready-made building blocks of the common network technologies and devices. Setting up an accurate representation of the network under consideration - together with appropriate traffic information - enables real life models to be built and run.

Note: The CACI site is used solely to illustrate the kinds of practical tools available and as a source of further background information. It does not mean that we necessarily endorse them in any way.

| Rules of Thumb | Analytical Techniques | Simulation | Benchmark |