A comparison of alternative Web service allocation and scheduling policies

Web services (WSs) are emerging as the building block of Internet scale database management systems (IDBMSs). These systems must intelligently execute plans that reference autonomous WSs. This requires policies and mechanisms for both scheduling and allocating WSs that constitute a plan. In this study, we analyze two scheduling strategies and four allocation policies. Obtained results show that dynamic scheduling with least response time (LRT) allocation policy is superior to other alternatives when the service time of a WS can be estimated accurately. The traditional least recently used (LRU) allocation policy is inferior to all policies including random. These observations are important because they impact the scalability of a system. Only with a smart allocation policy, one should expect improved system performance by increasing the number of nodes that constitute an IDBMS to support a larger number of WS replicas.