As a potentially revolutionary storage technology, solid-state drives (SSDs) promise superior speed and reliability over conventional hard disks. However, for some data experts, such conclusions remain questionable. Numerous companies have leapt to the answer the call for high-capacity solid state racks, and this brings about a serious need for credible and unbiased reports that accurately measure solid-state drive behavior.
The Time Machine: Early Approach to Measuring Solid-State Storage Performance
This beginning is based upon the measurements detailed in a paper produced by Daniel Myers and Matt Sinclair in a Fall 2010 report present to the University of Wisconsin-Madison. Without detailing every aspect of the tests performed by Myers and Sinclair, this review touches on the primary goals and then reveals the results.
In order to grasp the performance nature of solid-state storage technology, Myers and Sinclair based their tests upon two commercial SSDs. The fundamental workload revolved around randomized 4K writes across the full address space of each test drive. By including simulation and analytic modeling, the investigators also studied the effects of the flash translation layer (FTL), including garbage collection, wear-leveling and mapping processes.
The research focused on a limited range of SSDs. By using only NAND flash drives, the investigators leave room for future reports on NOR flash drives. Basic conclusions at the time of the study are as follows:
- The subject solid-state drives proved to be faster, more reliable and more energy efficient than disks
- The initial performance success of the subject SSDs proved to be lacking in the drive’s steady-state behavior
- The subject SSDs proved to be more costly per byte
- The subject SSDs had a limited number of write-erase cycles.
In the final analysis, the Myers and Sinclair studies revealed that 97 percent of the writes occurred in less than 300 µs, however the remaining writes reduced the mean write time to 13ms.
Back to the Time Machine: Current Approach to Measuring Solid-State Storage Performance
Times change and technology improves. Solid-state storage is still based upon the same principles that existed in 2010: Data is stored on silicon chips rather than mechanical devices such as streaming tape drives or high-speed metal platters. However, the capacities, speed and reliability of RAM and Flash chips have greatly improved.
According to Solid State Storage Initiative (SNIA), “Enterprise-grade solid state storage can achieve input and output data rates far greater than conventional, magnetic storage devices such as hard disk drives.”
Yet without some form of widely accepted testing methodology, such statements, even from reputable companies such as SNIA, can be biased and self-serving. This possibility automatically puts prospective buyers on guard and leaves them with a suspicious state of mind.
The work of individuals like Myers and Sinclair served as a temporary solution to a delicate problem. But now, the industry has begun to develop a reliable and industry-accepted methodology for measuring SSD performance. Buyers are finally able to make educated purchasing decisions based upon universal performance specifications without regard to manufacturer or cloud storage representatives.
Standards developed via the Solid State Storage (SSS) technology infrastructure define data storage processes, connective routes and the Flash chip interface to software interactions. By establishing a suite of tests and approved test methodologies, the SSS PTS defines the performance characteristics of SSDs.
This levels the playing field, enables fair-minded competition and establishes a means of measuring SSDs according to comparative Enterprise and Client SSD environments. For consumers and futurists alike, this means that a once potentially revolutionary storage technology has finely come of age.
