Model-Based Resource Provisioning

Publishing Venue

IBM

Abstract

A framework for utilizing workload models to predict resource requirements for adaptive provisioning in a hosting environment.

Model-Based Resource Provisioning

Past systems for resource provisioning and allocation of resources among competing services use a reactive system for determining when to make resource adjustments. Client response times are monitored and adjustments are made when thresholds are passed (indicating that an SLA may be violated). This invention describes a proactive approach based on predictive models of resource behavior, so that decisions can be made without the need to monitor client response time and before SLA's may have been violated.

One of the advantages to the model-based approach is that it also allows the provisioning system to represent complex inter-relationships between server clusters, storage subsystems, and grids. Our approach is model-based in that it incorporates simple analytical models of service behavior as a basis for predicting the performance of candidate resource allotments under changing load. The premise of model-based resource provisioning is that network service loads have common properties that allow the utility OS to predict their behavior. In particular, service loads are streams of requests with stable average case behavior; the model allows the system to adapt to changing resource demands at each stage by continuously feeding observed request arrival rates levels to the models to predict resource demands at each stage. Moreover, the models enable the system to consider interactions among stages in a comprehensive way, by predicting...