Explicit Control of Service Execution to Support QoS-Based Grid Scheduling

Abstract

Grid scheduling is shifting from a system-centric approach towards a user-centric one, i.e. service provision is driven by both user and provider-dependent Quality-of-Service (QoS) requirements. In this scenario, the possibility to explicitly control the execution of services allows providers to make different decisions on the QoS they provide their services with according to the requirements of new service requests. In the present work an infrastructure that allows providers to dynamically adapt the execution of services according to both the changing conditions of the environment where they operate in, and the requirements of service users is presented. The infrastructure is based on program continuations to provide service schedulers with application-level primitives to handle suspension and resuming of service execution. The same primitives are also accessible as web service operations by consumer programs so allowing to change QoS parameters of requested services at run-time. This approach makes the proposed infrastructure a flexible and easily programmable middleware to experiment with different scheduling policies in service-oriented scenarios. As a case of study, we show that on top of a time-sharing low-level scheduling, a provider can adopt a high-level scheduling policy using service suspension and resuming primitives so that consumer priority requirements can be met at run-time.