Section: New Results
Modelling and control of web servers
This work focuses on the design of a database server model by a fluid flow approach, taking into account elements of computing systems theory. This work is in collaboration with Sardes team of the INRIA.
The aim is to provide a model which could describe the behaviour of a server whatever may be the load. A first model based on a two tank system was developed using the fluid conversation law and the Little law. This model is relatively simple and is a first order system with time constant and gain depending upon the input of the system - that is the load of server [Oops!] , [Oops!] . We are now able to predict the output of the server through a model with self adaptive parameters. This model will then be used to build control laws to avoid the trashing phenomena that occurs in case of overload using admission control. The trashing phenomena is a nonlinear phenomena that increases the latency and decreases the performance of the server due to the increase of the multiprocessing level. We are now able to predict the output of the server through a model with self adaptive parameters.