Erlang C FAQ

Table of Contents

1. What are the assumptions underlying a valid Erlang C model?
2. Do I really need to use Erlang C for accurate answers?
3. How do I account for abandoned calls?
4. But my service times (talk and wrap) are approximately constant.
5. Where can I find out more about Traffic System design.

What are the assumptions underlying a valid Erlang C model?

The following assumptions underlay the Erlang C formula:

Do I really need to use Erlang C for accurate answers?

The number of servers required will always be large than the server traffic intensity. At quite large server traffic intensity (say 1000E) you may only need 1.02*server_intensity*1.02 servers to deliver good service, but at low intensity (say 1E) you may need 3*server_intensity servers to deliver good service.

How do I account for abandoned calls?

To make an allowance for calls that abandon quite soon after connection, just reduce the incoming call rate. You can do the same for calls that abandon later in long wait time models, but it is less legitimate. Nevertheless the inaccuracy introduced is probably small compared to the error bounds on the basic model parameters.

But my service times (talk and wrap) are approximately constant.

Erlang C will give conservative answers, it will over estimate the servers required. Usually the error is not very significant and is safe to Service Levels.

It is rare that queues don't occasionally overload, measurements during high load conditions provide a basis for calibrating an Erlang C model of your workload. You might try for instance to calibrate the model by compensating for abandoned calls.

Where can I find out more about Traffic System design.

There are textbooks available, although they not likely to be on the shelf in your local library.

You will also find info on the net, try your favorite search engine. You probably already know this, it is probably how you got here.

I have written a little on Traffic System design and provide handly calculators of some of the more frequently used functions, see my Telecommunications page.

Erlang C will give conservative answers, it will over estimate the servers required. Usually the error is not very significant and is safe to Service Levels.

It is rare that queues don't occasionally overload, measurements during high load conditions provide a basis for calibrating an Erlang C model of your workload. You might try for instance to calibrate the model by compensating for abandoned calls.

Return to Telephony Traffic Modelling