Context Aware Datacenter Load Balancer

Abstract

The ever increasing demand for cloud adoption is prompting researchers and engineers around the world to make the cloud more efficient and beneficial for all the stakeholders that include cloud service providers and cloud service users. Cloud computing will bring profits for all when the cloud resources are used efficiently, and its services are made affordable for businesses by reducing its cost. Managing cloud data center incurs a high cost, which includes capital expenditure for procuring necessary IT infrastructure at the beginning and recurring operational expenditures for data center management which includes power, manpower and maintenance. Data center owners need to reduce the data center management cost by employing efficient resource provisioning techniques to save energy and reduce cost without affecting the service level agreements. Load balancing is one of the critical aspects of cloud data centers that can significantly improve resource utilization, performance, and save energy by properly assigning/reassigning computing resources to the incoming requests. Therefore, how to schedule user tasks to virtual machines and virtual machines to physical servers effectively by considering various dynamic parameters is an evolving research problem in cloud computing. The proposed work investigates contextual parameters such as physical machine characteristics, data center load conditions, and electricity prices in the geodistributed data center locations to propose energy and cost-efficient load balancing technique for cloud data centers. The physical machine characteristics such as performance to power consumption profile are utilized for virtual machine placement decisions in data centers to optimize total energy consumption and improve throughput. The context of peak and non-peak load conditions is used to avoid virtual machine iplacement optimization overheads and efficient utilization of power-efficient physical servers. The electricity price varies according to geographical locations throughout the globe. The electricity price, along with response times, is considered to distribute data center loads optimally in geo-distributed data centers to save total power costs. Proposed work also investigates current challenges for efficient graphical processing units resource utilization in virtualized environments. The work proposes a context-aware load balancing technique that ensures better power-efficient resource utilization, enhances performance by avoiding overheads, and also saves total power costs of the data centers. The experimental results indicated that our proposed context-aware load balancer helps to save around 2-10% of power for synthetic workloads and 1-3% for real workload traces in the data centers. The experimental results also attested that our proposed cost-aware cloud service broker load distribution technique for geo-distributed data centers can save around 15-23% of power costs of the data centers.