next up previous contents
Next: Recurrent Self-Organizing Map Up: Computational Information Processing Previous: Learning Control of Indoor

Internet Services Modelling

Researchers: Jukka Heikkonen, Timo Koskela, Jouko Lampinen, and Kimmo Kaski

Project homepage

Internet Services Modelling project started in summer 1998. First phase of the project lasts until April 1999, and the second phase is scheduled to last two years. Project is funded by Technology Development Centre (TEKES) and Ministry of Transport and Communications. Research is conducted in collaboration with Lappeenranta University of Technology (LUT), Center for Scientific Computing (CSC), Nokia Research Center and Sonera.

Internet is very fast growing public network with now about one hundred million users worldwide. From the number of users point of view, Internet is quickly approaching the other two global public networks, the fixed and mobile telephone networks. The usage behaviour and the quality of service (QoS) on Internet, however, is not so well characterized and defined than on those other public networks. While there have been tens of years experience of measuring the traffic load during the peak hours and using different kinds of error rate behaviour and service availability measures, these are lacking on the Internet. The big problem in Internet is poor QoS. In order to know more about the QoS and in order to measure the Internet behaviour, it should be important to have some suitable Internet service model.

Project seeks for the generic model for classifying and modelling the characteristics of the Internet services from the user point of view. To achieve this, QoS must be defined in different usage situations.

Figure 9
Figure 9:   Elements of Internet service from the user point of view.

Figure 9 shows a schematic picture of the typical elements of the Internet service from the users point of view. User connects to Internet service provider's (ISP) server using a modem line or local network connection. Connections outside ISP's network are routed through proxy cache. The actual service that is being used is located on some distant server and is connected through the Internet. Experienced QoS is affected in each of the elements shown in the figure: the user itself (including client program, connection to ISP), ISP (modem lines, local servers and services, connections to Internet), proxy cache (cache size and configuration), Internet (network between the local and distant server) and the server connected (service or application used). Each of the elements also has a different set of QoS measures, that can also be dependent on each other. For example in the network element QoS can be measured e.g. as network throughput, latency time, transfer delay or percentage of dropped data packets. For other elements QoS may not be as easily measured, and consequently statistical or probabilistic models are used instead.

In the final phase different elements and the models of their QoS are brought together. This generic model can then be used as a simulator to predict the QoS in different situations and cases. Results are validated by comparing the predictions to the perceived QoS.

next up previous contents
Next: Recurrent Self-Organizing Map Up: Computational Information Processing Previous: Learning Control of Indoor