Pricing and autonomic control of web servers with time-varying request patterns

Abstract

The motivation for this paper stems from two important emerging trends: (i) the growth of commercial services on the Internet, and (ii) the need for deploying systems that can automatically tune themselves. The need for automatic tuning for price (charged by the service provider) becomes particularly acute not only because of high costs of involving humans in system operation but also because such tuning should be nimble enough to adapt to workload changes. This paper introduces a dynamic pricing structure which is fair to both clients and servers. It also examines techniques for dynamically setting the price. During widely time-varying loads, if the server charges a fixed price regardless of load, it can result in either being unfair to the clients or a loss in revenue earned. Again, fixing the price to a low value can decrease the revenue that a server could potentially make. Our first contribution in this paper is a combination of pricing and admission control strategy for such commercial services that take the above-mentioned factors into consideration. In the first step, the customer/client is offered a price package. The second step enforces a level of admission control imposed by the server. It can so happen that a client despite the high price, chooses to get service at a time of high load. The server can use the second step as a way of rejecting this client if it feels that the revenue it would make from this client can be offset by the lower revenue due to degradation in QoS for others. The second contribution in this paper is in developing and evaluating three heuristics for dynamically modulating the above pricing/admission-control strategy to maximize revenue. Some of the factors that need to be taken into consideration when designing techniques for modulating the parameters are revenue enhancement (for the server), implementability and computational complexity. We specifically focus on a Web service on the Internet that needs to cater to the http file requests of a large number of clients on a continuous basis.