Business Library

What about WAN?

We thought that it would be interesting to wrap up the year by looking briefly at a completely different area of networking. Suppose you have a business with a number of locations which require a permanent connection to a separate server farm location: How do you decide how to connect the sites together economically?

No surprises for guessing that what you don't do is try and figure something like this out by hand (unless the number of locations is very small) - you use a mathematical tool. Basically the problem is one of identifying where the locations are, what the cost of connecting each of them together is (using whatever line or services you wish to consider), and selecting certain locations to act as concentration points.

This demonstration focuses on a small subset of a real problem: The locations are positioned using screen coordinates; the "cost" of connecting each pair together is calculated as the Euclidean distance between them (see formula). The destination of all traffic is a single node, known as the "Centre", and the objective of the exercise is to cluster locations together so that the total line distance is reduced.

You should be looking at a panel on the left which is displaying 50 Blue spots and 1 Red one? If not, but you can see the row of push-buttons, press "Reset" and the picture should appear. If you cannot see anything, then your browser either doesn't support Java - or you have disabled it - in which case you won't be able to run this demo.

The panel represents a map of 50 locations of computer equipment which requires a connection to the Centre (the server farm) shown in RED. One option is of course to connect directly all the sites to the Computer Centre - and this is the starting point for the three heuristics (algorithms which do not guarantee an optimum answer ) demonstrated here: Each runs in a loop selecting which locations should act as intermediaries, known as Concentrators, for some of the other sites in order to reduce the total communications line cost. The heuristics are described below, together with the action to take for each step:

1. Add - where you start with no concentrators and add the one which saves the most.
2. CoM - which stands for "Centre of Mass" and starts with all locations as concentrators, then removes the one that saves the most - but positions the new concentrator at the centre of the cluster of locations formed - not at an actual location.
3. Drop - where you start with all locations as concentrators, and remove the one which saves the most. In this case, the dropped concentrator is clustered to another location which is still a concentrator.

In each case, the heuristic continues until no further cost savings are possible.

To run the demo, press one of the three buttons "Add", "CoM", or "Drop" and the program will produce a solution and show the reduction in distance made using the heuristic selected. "New" will produce a different set of locations, and as mentioned before "Reset" will redraw the starting network. Finally, the controls "Fast" and "Slow" control how quickly the demonstration runs.

You will see the following colour code used to make the map easier to read:

• Red - Centre
• Yellow - Concentrator
• Blue - Terminal (i.e. a location which is not a concentrator)

Finally. the results shown at the bottom of the map are:

• Concs - the number of concentrators selected
• Cost - the final (or current) total line distance
• Initial - the initial cost of connecting all locations to the Centre
• Saving - to save you doing some maths,=( Initial - Cost)

"Yes - but how close is this to real life?" you are thinking. The answer is - not as far away as you might imagine. If you ever use a practical WAN design tool, you will see much more complexity (such as coordinate and line tariff databases) but you should be able to recognise heuristics, like those used in the simple demo above, hiding under the covers.

If you wish to learn more about such techniques, Aaron Kershenbaum has written an excellent book called "Telecommunications Network Design Algorithms" which is now available from the Local Links Bookshop.

Business Implications?

• If you have a problem such as this to tackle in real life and haven't done anything similar before, you should probably look for specialist help. Only the largest organisations can justify maintenance of expensive and complex network design tools in-house - plus the supporting database (tariffs, locations, equipment etc.).
• Although design tools will give a solution every time, humans are still needed to check the quality of the answer - both technically and from a business perspective. This means that a good design is one which takes both aspects into account - and any design specialists involved are just one of the team in arriving at a practical design. If they won't do this, find someone else!
• The design tool is exactly that - a tool and only a tool.
• Design tools need to incorporate a lot of information to produce useful [outline] designs - but a few graphical representations, produced relatively quickly, aid the understanding of the relationship between the network specialist's design options and the underlying business requirements.

Feel free to experiment. Please e-mail wan@vinntec.co.uk if you wish to provide any feedback on this article or the program it contains.