Resource Allocation for Heterogeneous Cloud Computing Using Weighted Fair-Share Queues

摘要

On-demand resource provisioning is based on automated approaches for resource pooling and elasticity on the cloud service provider (CSP) side. The infrastructure services must be adapted dynamically to accommodate customer demands and yet, operate within offerings of the CSP. Although multiple approaches for homogeneous clouds are available, more realistic platforms based on heterogeneous resources and virtual machines (VMs) present unique challenges. Our resource management algorithm allocates memory, network and computational resources to heterogeneous VMs, in order to provide customized fine-grained control for the scalable capacity planning of data centers. Weighted fair-share (WFS) queues: high, medium and low are used to classify the incoming jobs in buckets of appropriate length based on priority. The highest priority jobs with aggressive deadlines are allowed to progress at a similar pace using round-robin scheduling, while lowest priority jobs are allocated on Low Queue with First-Come-First-Serve (FCFS) scheduling. The proposed algorithm performs better on throughput related metrics: number of instructions executed (30% more), turn-around and waiting times (on an average of about 10% less) w.r.t. standard policies such as shortest job first (SJF) and FCFS.