Optimization Problems in Service Provisioning Systems

A service provisioning system typically contains a number of servers which may be distributed, heterogeneous and intermittently unavailable. They are used by the host in order to offer different services to a community of users. There may or may not be Service-Level Agreements involving Quality of Service constraints. In this context, there are several areas where dynamic optimisation problems arise quite naturally. These are (a) Routing and load-balancing: Where should an incoming request be sent for execution? If some queues grow large while others are short, can something be gained by transferring jobs among them? (b) Resource allocation: If different servers are dedicated to different types of service, how many should be assigned to each? When should a server be switched from one type of service to another? (c) Revenue maximisation:How are resource allocation and job admission policies affected by economic considerations? In particular, if service-level agreements specify payments for serving jobs and penalties for failing to provide a given quality of service, how many servers should be assigned to each type of service and when should jobs of that type be accepted?