Optimization of hot/cold separation algorithm computation for SSD
TimeMonday, July 11th10:45am - 11am PDT
Location2012, Level 2
Event Type
Embedded Systems and Software
Engineering Tracks
Embedded Systems
DescriptionIt is well known that being able to properly classify the hotness of data on an SSD has a great advantage in performance. There are two main methods for classifying hotness of data. 1) the method of classifying based on access frequency 2) the method of classifying based on sector size. In the case of this paper, any pattern or trace the calculation amount of access frequency method that can be accurately classified was optimized. This paper introduces a method of optimizing WDAC (Window-based Direct Address Counting) among them. WDAC method had different weights of information for each incoming command, so the consistency of information was excellent through the recency information. However, every time a command enters the SSD from the host, by calculating the weight of all the addresses in the window, there was a big drawback of computational overhead. Additional information can be used to determine how to solve the shortcomings.