Advisor(s)

Ningfang Mi

Contributor(s)

Yunsi Fei, Kaushik Chowdhury

Date of Award

12-2012

Date Accepted

12-2012

Degree Grantor

Northeastern University

Degree Level

M.S.

Degree Name

Master of Science

Department or Academic Unit

College of Engineering. Department of Electrical and Computer Engineering.

Keywords

multi-tiered storage system, performance degradation, SSD

Disciplines

Data Storage Systems | Electrical and Computer Engineering | Engineering

Abstract

With the fast development of the semiconductor industry, solid-state disk (SSD) has been more and more widely used in storage systems. How to take better advantage of SSD devices in storage systems gradually becomes a hot topic nowadays. Although the price for SSDs decreases in a fast pace and becomes acceptable in a variety of applications, the gap in price per Gigabyte between SSDs and traditional hard drive devices is still tremendous. As a result, multi-tiered storage systems which combine both SSDs and hard drive disks (HDDs) as their storage devices are developed for finding the best balance between cost and performance, i.e., achieving the performance as good as SSDs and meanwhile maintaining comparatively low cost. In order to develop such an efficient multi-tiered storage system, a thorough understanding of both SSDs and HDDs is highly needed. Comparing to traditional hard disks, solid-state disk drives are more complicated with a large number of features and mechanisms that are still unclear. Therefore, in this paper we first study the performance degradation under different I/O access patterns. After observing the phenomenon of non-negligible performance degradation when we have continuous random writes on the SSD, we develop two models: a theoretical model called Pump-Pool model to capture the performance degradation as a function of the SSD occupancy rate, and a simulation model of a multi-tiered storage system in which performance degradation in the SSD tier is precisely described. We finally design a new data migration algorithm which dynamically monitors the application workloads and changes the system's configuration to achieve better I/O bandwidth in an adaptive way. In additional, we take endurance degrading rate into consideration in our algorithm, and then we conduct a set of trace-driven simulations to evaluate the new algorithm. The simulation results show that good performance improvement in terms of both speed and endurance is achieved under our new algorithm.

Document Type

Master's Thesis

Rights Holder

Cai Wei


View Full Text

Click button above to open, or right-click to save.

Share

COinS