New Trends In Computer Networks

KATJA GILLY AND SALVADOR ALCARAZ
Universidad Miguel Hern ndez, Departamento de F sica y Arquitectura de Computadores, Avda. de la Universidad, s/n, 03202 Elche (Spain). Email: {katya,salcaraz}@umh.es
CARLOS JUIZ AND RAMON PUIGJANER
Universitat de les Illes Balears, Departament de Ci ncies Matem tiques i Inform tica, Ctra. de Valldemossa, km 7.5, 07122 Palma de Mallorca (Spain). Email: {cjuiz,putxi}@uib.es
Prediction is one of the most important factors that prevents a Web system from collapsing when high intensity of transactional traffic arrives to it. It is necessary to ensure the quality of prediction is good enough in order to maintain under control the correct utilization of the Web system by allocating its resources among different types of incoming traffic that can reach the system and by controlling bursty transaction arrivals to the system. We have developed an algorithm that resides in a Web switch and distributes the workload among a set of servers following a resource allocation policy based on quality of service (QoS) attributes. The algorithm includes a reward scheme to control the accuracy of the predictions on real time.
The Internet community is continuously growing and the need of performance studies in this field is becoming crucial to obtain better service response times for the web users. Within the parameters to be controlled in order to avoid web system congestion, some of them are related to traffic arrival attributes at web servers. If the web system is receiving arrivals that follow a constant distribution, then the situation is simplified considerably. Unfortunately this...