Reinforcement Learning routing
Author Image

By: Aivis Olsteins

In News


VoIP routing part: 4. Least cost routing and beyond

LCR, Quality and priority based routing models may not be enough to build an advanced routing system for VoIP.

Previously we discussed several routing modes most frequently used in VoIP telecoms: priority based routing, least cost routing and quality based routing. While these methods seem to be covering all cases and allowing to build comprehensive routing system, we can show you that it is not the case.

First of all, how do you choose when to use which model? Common argument here is that in most cases there are relatively small number of routes for which given operator has most of the traffic, so called 80% rule (i. e. 80% of traffic is to top 20% of destinations and vice versa). In such cases top 20% destinations can be managed manually, i. e. routes can be set manually knowing each routes perceived performance. The rest 80% is normally set to least cost to ensure profit margins stay good enough without doing too much manual work by configuring each and every destination individually. This model works up to the point, until some of the top 20% route hits quality issue. The system will not reroute itself in case of such failure and would require manual intervention from the operator. Setting these routes to quality only routing would also not work, because since these routes provide most of the income (top 80% of traffic), following the quality changes might lead to unpredictable profit margin changes. Same goes to the rest 80% of routes which are set to LCR: the quality fluctuations there would probably go unnoticed and may lead to degradation of the service on a larger scale.

Above example illustrates the need for more general routing model which could incorporate more that quality or price only factors. Ideally the system should take large range of the parameters, historical data of the performance and based on that build the expected model to determine

The list of factors which might affect the selection of most optimal route can include a list of parameters:

  • Cost of the specific route (the case of least cost);
  • ASR (answer/seizure ratio);
  • ACD (average call duration);
  • PDD (post dial delay);
  • number of active calls to the current provider/carrier (helps to detect possible overloading of the carrier)
  • rate of call attempts per second (to detect possible
  • time of day / day of week (to detect daily pattern change)
  • time if the day at the destination (take in account time-of-day pattern variations)
  • other data?

The routing model should be able to take the above parameters as the input and create data model which can feed the routing engine to select a route which will lead to the best possible result.

Also we should define what is the best possible result. There can be several criteria which can serve as the indicators to the best possible result, like:

  • highest total number of minutes processed by the system (i. e. highest ACD x ASR product)
  • highest maximum profit generated, i. e. this takes in account both: the highest number of minutes processed (maximum revenue) and minimum cost (least cost).

The final choice should be decided by the system operator, and the routing system should be able to work with any of the criteria for best possible result.

Leave a comment
About Author
Author Image
My name is Aivis Olsteins and I am owner of DataTechLabs. My experience in Telecoms started in early 1990's and I have worked in multiple technical positions in mobile, messaging and data networks. My expertise lies in telecom networks, database systems, distributed processing and large data analysis. These posts are my attempt to share my knowledge with everyone who might find it useful.

Get in Touch

If you have something to say, please use the contact form below to get in touch with us. We will get back to you as soon as possible.

Mail Us

DataTechLabs SIA, Muzikas str 12A
Jurmala, LV-2008, Latvia.

Call Us

+371 67 66 09 01
+ 1 202 499 1550

E-mail Us