Multi-constraints Dynamic Virtual Resource Management In Clouds

In this paper, we addressed the new challenges raised bymulti-constraints resource management for different shareholders. Multipleconstraints partly come from cloud users who specify the qualityrequirements in Service Level Agreement (SLA) with cloud providers,such as availability and performance requirements. While other multipleconstraints come from cloud service providers, such as saving operatingcosts by reducing energy consumption in order to achieve high profits. Theresource management activities should balance the benefits of differentshareholders. For example, allocating more virtual resources for cloudapplications can ensure the satisfaction of performance and availabilityrequirements to some extent. However, it will increase the operating costsof cloud providers. On the contrary, allocating less virtual resources forcloud applications can effectively reduce the operating costs for cloudproviders, but may lead to the violation of SLA terms on response time andavailability. So how to allocate, place and consolidate virtual resources tomeet multiple constraints from different shareholders has become acommon concern for both cloud users and cloud providers.Therefore, we proposed a multi-constraints virtual resourcemanagement framework by separating different concerns of cloud usersand cloud providers and satisfying them in different levels. Cloud resourcemanagement is divided into the application level and the platform level.Specifically, in the application level, the mapping process from applicationquality constraints to virtual resources is abstracted as the applicationcontroller. Performance requirements are mapped to the quantity of virtual resources and the availability requirements are mapped into the placementplan with resource locations. In the platform layer, virtual resources areconsolidated into the optimal number of physical servers to minimizeenergy consumption and operating costs.In the application level, we focused on the mapping process ofavailability requirements to resource placement. An availability-awareplacement approach to dynamic scaling is provided based on anavailability computation model for cloud applications. It performed twotypical scaling strategies—vertical and horizontal resizing—to generate theplacement plan. The vertical resizing adjusts the partition of resourceinside virtual machines while horizontal resizing changes the number ofvirtual machine instances. Since the two resizing strategies will havedifferent impacts on the availability, communication overhead andsoftware license costs of applications, the approach chooses vertical orhorizontal resizing strategies depending on the difference between thecurrent availability and the desired one.In the platform level, we provided an improved resourceconsolidation approach to fulfill the energy-saving requirement for cloudproviders. The consolidation approach is composed of reactiveconsolidation and periodical consolidation. To be specific, reactiveconsolidation locally runs on each physical server, detecting anomaly andperforming consolidation autonomously. The physical servers of which theutilizations of computing resource exceed the upper limitation or fallbelow the lower limitation are marked as anomaly. In each case, weprovided a relocation algorithm to develop consolidation solutions.Periodical consolidation, as a complementary of the reactive one,periodically performs global consolidation for all physical servers in thesame availability zone.Finally, in order to evaluate the effectiveness of our approaches—theavailability-aware resource scaling placement approach and the improvedconsolidation approach, we designed and implemented two experiments by simulating a hierarchical cloud. The first experiment simulated two typicalscenarios—scaling up and scaling down—and compared the proposedplacement approach with horizontal-only and vertical-only resizingapproaches on four merits. We conducted the second experiment forresource consolidation in the same simulated cloud with multiple physicalmachines and virtual machines in different granularities. The results of twoexperiments showed that the proposed consolidation approach reducedmigration costs and was an effective strategy with a practical applicationvalue.In summary, we addressed the new challenges raised bymulti-constraints resource management for different stakeholders.Constraints from the different stakeholders were separated and met indifferent levels. Specifically, we focused on the resource placement issueto meet availability constants and resource consolidation issue to fulfillenergy-saving demands, in order to achieve the situation with mutualbenefits for cloud users and cloud providers.