HOTIS: A Hot Data Identification Scheme to Optimize Garbage Collection of SSDs

With the development of the computer hardware, storage devices have become the performance limit to the whole computer system. Solid State Disk (SSD) is a new kind of storage device which uses flash memory as the basic memory cell. It has much better random read and write performance than traditional Hard Disk Drive (HDD), and has gradually taken place of the traditional HDDs. SSD also has some special disadvantages, for example, garbage collection(GC) process. To overcome these disadvantages, several methods are proposed to reduce the frequency of GC or the time penalty caused by GC. One way is to separate the hot and cold data and write them into different blocks, which can decrease the number of page migration. However, existing hot data identification methods are derived from cache algorithms such as LRU and LFU. Thus, they have ignored the differences between hot data in SSD and cache, as will as the special features of SSD. To address this problem, we proposed a hot data identification scheme called HOTIS. HOTIS exploits the information collected during garbage collections, and calculates the number of blocks with fewer valid pages to dynamically set the threshold of IRR(Inter-Reference Recency). To demonstrate the effectiveness of HOTIS, we conduct several simulations via Disksim. The evaluations show that, compared to traditional policy, HOTIS reduces the number of page migrations by up to 92.40%, and decreases the average response time by up to 8.47%.