Cloud Workload Characterization
Wira D. Mulia, John M. Acken, D. Fritz, Naresh Kumar Sehgal, Sohum Sohoni, CLucas Stanberry · IETE Technical Review · 2013
In this chapter, we describe various Cloud workloads and optimization issues from the points of view of various players involved in Cloud Computing. A comprehensive categorization of various types of diverse workloads is proposed, and nature of stress that each of these places on the resources in a data center is described. These categorizations extend beyond the Cloud for completeness. The Cloud Workload categories proposed in this chapter are: Big Streaming Data, Big Database Creation and Calculation, Big Database Search and Access, Big Data Storage, In-Memory Database, Many Tiny Tasks (Ants), High-Performance Computing (HPC), Highly Interactive Single-Person, Highly Interactive Multi-Person Jobs, Single-computer intensive jobs, Private Local Tasks, Slow Communication, Real-Time Local Tasks, Location-Aware Computing, Real-Time Geographically Dispersed, Access Control, and Voice or Video over IP. We evaluate causes of resource contention in a multi-tenanted data center and conclude by suggesting several remedial measures that both a Cloud Service provider and Cloud customers can undertake to minimize their pain points. This chapter identifies the relationship of critical computer resources to various workload categories. Low-level hardware measurements can be used to distinguish job transitions between categories and within phases of categories. This relationship with the categories allows a technical basis for SLAs, capital purchase decisions, and future computer architecture design decisions. A better understanding of these pain points, underlying causes, and suggested remedies will help IT managers to make intelligent decisions about moving their mission-critical or enterprise-class jobs into Public Clouds.