Cloud computing is a large-scale distributed computing paradigm driven by economies of scale, in which a pool of abstracted, virtualized, dynamically-scalable, managed computing power, storage, platforms, and services are delivered on demand to external customers over the Internet. Considering the lack of resources in cloud environments and fluctuating customer demands, cloud providers require to balance their resource load and utilization, and automatically allocate scarce resources to the services in an optimal way to deliver high performance physical and virtual resources and meet Service Level Agreement (SLA) criteria while minimizing their cost.
This study proposes an Autonomic System for Optimal Resource Management (AS-ORM) that addresses three main topics of resource management in the cloud environment including: (1) resource estimation, (2) resource discovery and selection, and (3) resource allocation. A fuzzy Workload Prediction (WP) sub-system and a Multi-Objective Task Scheduling optimization (MOTS) sub-system are developed to cover the first two aforementioned topics. The WP sub-systems estimates Virtual Machines’ (VMs’) workload and resource utilization, and predicts Physical Machines’ (PMs) hotspots. The MOTS sub-system determines the optimal pattern to schedule tasks over VMs considering task transfer time, task execution cost/time, the length of the task queue of VMs and power consumption.
To optimize the third topic in resource management, resource allocation, VM migration that is the current solution for optimizing physical resources allocation to VMs and load balancing among PMs, is investigated in this study. VM migration has been applied to system load balancing in cloud environments by memory transfer, suspend/resume migration, or live migration for the purpose of minimizing VM downtime and maximizing resource utilization. However, the migration process is both time- and cost-consuming as it requires large size files or memory pages to be transferred, and consumes a huge amount of power and memory for the origin and destination PMs especially for storage VM migration. This process also leads to VM downtime or slowdown. To deal with these shortcomings, a Fuzzy Predictable Task-based System Load Balancing (FP-TBSLB) sub-system is developed that avoids VM migration and achieves system load balancing by transferring extra workload from a poorly performing VM to other compatible VMs with more capacity. To reduce the time factor even more and optimize load balancing over a cloud cluster, FP-TBSLB sub-system applies WP sub-system to not only predict the performance of VMs, but also determine a set of appropriate VMs that have the potential to execute the extra workload imposed on the poorly performing VMs. In addition, FP-TBSLB sub-system employs the MOTS sub-system to migrate the extra workload of poorly performing VMs to the compatible VMs.
The AS-ORM system is evaluated using a VMware-vSphere based private cloud environment with VMware ESXi hypervisor. The evaluation results show the benefit of the AS-ORM in reducing the time taken for the load balancing process compared to traditional approaches. The application of this system has the added advantage that the VMs will not be slowed down during the migration process. The system also achieves significant reduction in memory usage, execution time, job makespan and power consumption. Therefore, the AS-ORM dramatically increases VM performance and reduces service response time. The AS-ORM can be applied in the hypervisor layer to optimize resource management and load balancing which boosts the Quality of Service (QoS) expected by cloud customers.