Degree Type

Thesis

Date of Award

2014

Degree Name

Master of Science

Department

Electrical and Computer Engineering

First Advisor

Arun K. Somani

Abstract

Broader availability of solid state storage devices (SSD) enables an opportunity to improve the performance of the storage architectures. However, it not clear how exactly this development should proceed in terms of achieving greater cost-performance balance. Designers will benefit from a data-driven approach to understand how much SSD they should invest in to realize the most cost effective system.

This paper presents an analysis of actual workloads to deduce and derive guidance for an optimal investment strategy to balance the solid state and hard disk drive (HDD) to achieve the best cost and performance trade-offs. We show that while it is possible to determine this balance, it is heavily application dependent. For the workloads we studied, under certain assumptions, the preferred proportion of SSD varies from 8% to 60% for an 80% improvement in I/O performance (measured in terms of hits in SSD) compared to totally magnetic disks.

Further, we also propose three replacement strategies to keep the most accessed data in SSD. This replacement is determined using the past usage data. The goal is to make the best use of the available SSD, while minimizing the number of replacements. Our Simulation results shows that the best of our strategies provide 60% to 90% performance improvement compared to totally HDD across different workloads.

Copyright Owner

Joy Shukla

Language

en

File Format

application/pdf

File Size

38 pages

Download

Share

COinS