Abstract
In recent years, Internet-connected devices have become ubiquitous, leading to increased demand for computing and storage resources. The cloud has emerged as a cost-effective and scalable solution to infrastructure requirements, making it possible to consolidate several disparate computing resources into a single physical machine. However, while the availability of multicore CPUs, high-capacity memory modules, and virtualization technologies has increased the density of converged infrastructure, disk I/O has remained a substantial performance bottleneck, especially when dealing with high-capacity mechanical disks. Here, the authors investigate how proactive disk scheduling, backed by predictive models and client-side coordination, can influence the overall throughput and responsiveness of a cluster in data-intensive computing environments. They evaluate their framework with a representative MapReduce job on a 1,200-virtual-machine cluster, demonstrating a 21 percent improvement in completion time under heavy disk contention.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have