An adaptive admission control and load balancing algorithm for a QoS-aware Web system A dissertation presented by Katja Gilly de la Sierra - LLamazares to The Department of Ci`encies Matem`atiques i Inform`atica in partial fulfilment of the requirements for the degree of Ph.D. in the subject of Computer Science Universitat de les Illes Balears Palma de Mallorca, Spain September 2009 ii Dedicated to Lolo and to my parents iii iv Acknowledgments Now that this doctoral work is almost completed I am happy to say that it has been absolutely worth doing it. I have enjoyed most of the time I have been working on it over these last six years and I have learnt about many more things that I thought I was going to, like how to coherently write a paper, give a talk and work in a group of people, apart from staying focused and research in an area that is continuously changing. It is also true that, at some points, it has been very difficult to keep going, but the support and useful comments of my supervisor has always helped me to get rid of the bad “research” moments and look forwards. First of all, I want to thank Dr. Carlos Juiz for the patience, support and friendly supervision he has given me. His invaluable advice and stimulation has made this thesis possible. I also would like to thank Prof. Ramon Puigjaner, who has always given me kind advice and very interesting comments that helped me to improve this work. I also have to thank Salvador Alcaraz for the coffee discussions, comments and many jokes that helped me to keep my feet on the ground. I am very grateful to Dr. Nigel Thomas, who invited me on two occasions to visit the School of Computing Science at Newcastle University for three months. This has permitted me,apartfromimprovingmyEnglishandmeetingamazingpeople,tospeeduptheprogress of this thesis. And finally I want to thank Lolo, my partner and guide during all these years for his constant and tireless support and love during the good and bad moments. I would also like to thank my parents who have always given their love and encouragement. v vi Acknowledgments Thesis supervisor Author Dr. Carlos Juiz Garc´ıa Katja Gilly de la Sierra - LLamazares Abstract The overload of the servers and the resulting decrease in the Quality of Service (QoS) and performance becomes more serious as the use of Web services grows. In order to avoid this, service providers use large distributed networks of servers to attend the requests of the increasing number of visits in popular sites. The main objective of this thesis focuses on the design of an adaptive algorithm for admission control and content-aware load balancing for Web traffic. In order to set the context of this work, several reviews are included to introduce the reader in the background concepts of Web load balancing, admission control and the Internet traffic characteristics that may affect the good performance of a Web site. The admission control and load balancing algorithm described in this thesis manages the distribution of traffic to a cluster of Web servers (or Web cluster) based on QoS re- quirements. The goal of the proposed scheduling algorithm is to avoid situations in which the system provides a lower performance than desired due to servers’ congestion. This is achieved through the implementation of forecasting calculations. Obviously, the increase of the computational cost of the algorithm results in some overhead. This is the reason for designing an adaptive time slot scheduling that sets the execution times of the algo- rithm depending on the burstiness that is arriving to the system. Therefore, the predictive scheduling algorithm proposed includes an adaptive overhead control. Once defined the scheduling of the algorithm, we design the admission control module based on throughput predictions. The results obtained by several throughput predictors are compared and one of them is selected to be included in our algorithm. The utilisation level that the Web servers will have in the near future is also forecasted and reserved for each service depending on the Service Level Agreement (SLA). Ourload balancingstrategy isbasedonaclassical policy. Hence, acomparisonof several classical load balancing policies is also included in order to know which of them better fits our algorithm. A simulation model has been designed to obtain the results presented in this thesis. vii viii Abstract Director de tesis Autor Dr. Carlos Juiz Garc´ıa Katja Gilly de la Sierra - LLamazares Resumen La sobrecarga de los servidores y la disminuci´on resultante en la calidad de servicio (QoS) y rendimiento se agrava conforme la demanda de servicios Web aumenta. Con el fin deevitarlo, los proveedores deservicio utilizan redes distribuidasdeservidores paraatender el creciente nu´mero de visitas en sitios Web populares. El objetivo principal de esta tesis se centra en el disen˜o de un algoritmo adaptativo de control de admisi´on y equilibrio de carga para tr´afico Web. Con el fin de establecer el contexto deesta tesis, se realiza unadescripci´on exhaustiva delos conceptos de distribuci´on equilibrada de carga Web, control de admisi´on y las caracter´ısticas del tr´afico de Internet que pueden afectar el buen desempen˜o de un sitio Web, adem´as de su estado del arte. El algoritmo de control de admisi´on y equilibrio de carga descrito en esta tesis gestiona la distribuci´on del tr´afico en un clu´ster deservidores Web en base a los requisitos decalidad de servicio. El objetivo del algoritmo propuesto es evitar situaciones en las que el sitio Web proporciona un rendimiento m´as bajo de lo deseado debido a la congesti´on de servidores. Esto se consigue en base a c´alculos de predicci´on. Obviamente, el aumento del coste com- putacional durante la ejecuci´on del algoritmo supone una carga adicional en los recursos del sistema (overhead). Por esta raz´on, se ha disen˜ado una planificaci´on adaptativa en la ejecuci´on delalgoritmo quevar´ıaenfunci´ondelasr´afagas detr´aficoWeb quesedetectan en el sistema (burstiness). Por lo tanto, el algoritmo propuesto incluye un control adaptativo del overhead. Una vez definida la planificaci´on del algoritmo, disen˜amos el m´odulo de control de ad- misi´on basado en predicciones de productividad (throughput). Los resultados obtenidos por varios predictores de productividad se comparan y uno de ellos es seleccionado para ser incluido en nuestro algoritmo. El nivel de utilizaci´on de los servidores Web tambi´en se predice y se reserva para cada servicio en funci´on de compromiso adquirido (SLA). Nuestra estrategia de equilibrio de carga se basa en una pol´ıtica cl´asica de distribuci´on. Por tanto, seincluyeunacomparaci´on devarias deestas pol´ıticas ejecutadas enelalgoritmo con el fin de saber cu´al de ellas proporciona mejores resultados. Se ha disen˜ado un modelo de simulaci´on con el que se han obtenido los resultados que se presentan en esta tesis. ix x Resumen
Description: