Dynamic Provisioning of System Topologies in the Cloud

Abstract

Today’s IT infrastructures of companies are dimensioned to cover highly diverse workloads. In particular, it must be guaranteed that peak workloads can be processed according to concerted Service Level Agreements (SLAs). Consequently, companies have to cope with high acquisition costs of IT resources such as machines and software including costs for their maintainance and operation to meet these requirements. Thereby, achieving a high utilization of the available resources during most of the time is not possible. Based on these facts, companies endeavor to outsource their IT infrastructure to IT service providers, which in turn intend to offer respectively tailored and on-demand usable IT services using cloud computing paradigms. Obviously, the IT service providers are anxious to minimize the total cost of ownership (TCO) of their operating environments. Therefore, their goal is to minimize the amount of the provisioned IT resources by meeting tenant-specific SLAs and to maximize the utilization of the hosted IT resources by sharing them among multiple tenants (multi-tenancy). This paper presents a dynamic and cost-efficient provisioning approach of multi-tenant capable system topologies based on a Monitor-Analyze-Plan-Execute (MAPE) loop concept. For workload estimation and derivation of a capable resource topology, the MAPE loop is executed regularly regarding specified time intervals, which forms a proactive dynamic provisioning approach. Thereby, the proposed provisioning techniques apply heuristics which already encapsulate concrete performance information instead of using complex performance model solutions. Finally, a topology calculation model is developed which is the base for the proposed dynamic provisioning approach. This model enables provisioning capabilities supporting customer demands, cost-efficient utilization of resource instances, and sharing of resources by multiple tenants.