Understanding performance anomalies of SSDs and their impact in enterprise application environment

Abstract

SSD is known to have the erase-before-write and out-of-place update properties. When the number of invalidated pages is more than a given threshold, a process referred to as garbage collection (GC) is triggered to erase blocks after valid pages in these blocks are copied somewhere else. GC degrades both the performance and lifetime of SSD significantly because of the read-write-erase operation sequence. In this paper, we conduct intensive experiments on a 120GB Intel 320 SATA SSD and a 320GB Fusion IO ioDrive PCI-E SSD to show and analyze the following important performance issues and anomalies. The commonly accepted knowledge that the performance drops sharply as more data is being written is not always true. This is because GC efficiency, a more important factor affecting SSD performance, has not been carefully considered. It is defined as the percentage of invalid pages of a GC erased block. It is possible to avoid the performance degradation by managing the addressable LBA range. Estimating the residual lifetime of an SSD is a very challenging problem because it involves several interdependent and mutually interacting factors such as FTL, GC, wear leveling, workload characteristics, etc. We develop an analytical model to estimate the residual lifetime of a given SSD. The high random-read performance is widely accepted as one of the advantages of SSD. We will show that this is not true if the GC efficiency is low.