Autonomic Mechanisms for Service Scaling in Wide-Area Shared Computing Environments

In this paper we present mechanisms and models for building autonomically scalable and resilient services in wide-area shared computing environments. Wide-area shared computing platforms as exemplified by PlanetLab provide a cooperatively shared pool of computing resources. Building services in such environments poses unique challenges due to the fluctuations in available resource capacities and node availability. Autonomic mechanisms for building scalable and resilient services in such environments need to appropriately scale the service capacity under fluctuating available resource capacities, node crashes, and fluctuations in client load. Towards addressing these issues, we have developed models and mechanisms for estimating the service capacity under varying available resource capacities through online micro-benchmarking. The mechanisms for autonomic scaling of service capacity perform dynamic replication and regeneration of service replicas based on the estimated service capacity and observed load. This requires selection of appropriate nodes for the placement of new replicas. Furthermore due to the fluctuations in resource capacities, the load distribution for such services need to adapt to the varying service capacities of the individual service replicas. For this purpose we develop mechanisms for adaptive load distribution. We present here the evaluation of these mechanisms through experiments conducted over the PlanetLab environment.