Live Migration in Emerging Cloud Paradigms

VIRTUALIZATION, ELASTICITY, AND RESOURCE SHARING ENABLE NEW LEVELS OF FLEXIBILITY, CONVENIENCE, AND ECONOMIC BENEFITS, BUT THEY RESULT IN NEW CHALLENGES FOR PERFORMANCE AND PRIVACY, AS WELL AS NEW POTENTIAL SECURITY VULNERABILITIES. Security, privacy, and performance are major concerns for both public and private organizations that wish to shift their business-critical and sensitive data to the cloud. According to the existing cloud model, cloud customers use the services offered by cloud service providers (CSPs) without knowing exactly where their virtual machines (VMs) and data are stored or located and where they’ll be migrated. Such problems are exacerbated in emerging cloud paradigms, such as cloud federation and fog computing. Specifically, cloud federation potentially represents a new business model, in which different (geographically distributed) CSPs aggregate resources to create large-scale distributed virtual computing clusters, operating as though they’re within a single cloud organization. 1 This offers greater guarantees in terms of resilience and scalability required by data- and media-intensive critical applications. Fog computing is the evolution of cloud computing toward the Internet of Things, in that it extends the cloud’s elastic resource provisioning to the edge of the network, such as portable devices, smart objects, and wireless sensors.2 To ensure that customer expectations can be fulfilled, services offered in such emerging cloud paradigms should be formally defined by servicelevel agreements (SLAs) and total cost of ownership (TCO). An SLA is part of the service contract between a CSP and customer that specifies each party’s obligations and describes the desired quality of service (QoS) terms. As IEEE Cloud Computing Editor-in-Chief Mazin Yousif remarked, an SLA usually doesn’t include cost elements, which are determined through a separate pricing document. If the services listed in the SLA don’t meet customers’ expectations, the penalty is imposed on the CSP. Upon conclusion of the SLA negotiation and the brokering of the cloud resources, computing and storage resources must be allocated within the cloud infrastructure (Figure 1). This process can be modelled as an optimization problem of resource allocation with the objective of determining the allocation plan that minimizes the number of used physical host machines under the constraints of the requirements expressed in the SLA.3 Several metaheuristic approaches address this problem.4 However, while consumers are using the virtual resources, applications, and services, CPU load, memory capacity, and network traffic volume can increase. Replication, migration, and resizing represent different strategies and approaches for handling resource reallocation in the cloud infrastructure to provide the required scaling and elasticity capabilities.5 For example, in fog computing, computational tasks are executed close to the edge nodes of the network as needed. Specifically, user mobility and possible network disconnections require frequent migration of applications and data to meet end-to-end latency restrictions, preserve resources in user devices, and conserve bandwidth in the infrastructure.