Smart Resource Allocations for Highly Adaptive Private Cloud Systems
Abstract
Private cloud systems have the potential to offer the advantages of virtualization in terms of resource utilization to enterprises that can’t choose migrating their data and applications outside of their premises, for legal, privacy or compliancy reasons. However, in order to achieve their full potential, these private clouds need to be extended in order to support an integrated and adaptive behaviour in regards to the specific applications that are executed on the upper layers. Modern applications are more and more aware of the fact that the underlying platform is virtualized and so resources might be allocated and de-allocated in an adaptive fashion with respect to the current load and capacity. This paper presents a service oriented mechanism for adaptability of a typical private cloud system to work load fluctuations that is capable of intelligent resource allocation in both terms of amount and co-locations based on virtualization optimization. The real time monitoring information is gathered with a multi-agent system capable of multi-layer and multi-factor monitoring. The smart resource allocation is achieved with a distributed genetic algorithm that considers the workload characteristics in conjunction with physical optimum allocation and the current load. The pilot implementation is presented in the context of IBM Cloud Burst 2.1 private cloud implementation with a study on DayTrader J2EE benchmark application in load test scenarios. The results illustrate how the private cloud can show an adaptive behaviour related to the work load variations.