Download the authoritative guide: Enterprise Data Storage 2018: Optimizing Your Storage Infrastructure
By Teresa Worth and David Szabados
Solid state storage is an exciting technology area that offers the promise of high performance for a wide range of applications in the enterprise. Unfortunately, some SSD vendors have caused confusion with bold claims (and by positioning their products for the enterprise based on those claims) that have not always lived up to expectations during actual use. A lack of standards for the enterprise SSD market, and an application environment for businesses which cannot tolerate excessive failures, has compounded this problem.
Designing an SSD: The Basics
The design of an SSD involves two key areas that must be integrated for the final solution: the non-volatile memory and the controller.http://o1.qnsr.com/log/p.gif?;n=203;c=204655439;s=10655;x=7936;f=201806121855330;u=j;z=TIMESTAMP;a=20400368;e=i
The memory used most commonly in SSDs is non-volatile NAND flash memory and is produced in varying grades of quality. NAND is designed based on using either single-level cell (SLC) or multi-level cell (MLC) technology. SLC maintains one data bit per cell, and by nature has longer endurance, but SLC is significantly more costly to produce with higher capacities. SLC has a disadvantage of having high cost mixed with less overall capacity.
MLC NAND has lower endurance, as multiple data bits are shared in each cell, but MLC holds larger capacities and can be produced at much lower costs. MLC is offered in 3-bit per cell and 2-bit per cell varieties, with 3-bit per cell implementations having the highest capacity available at the exchange of slower performance.
After the NAND, a critical component of an SSD is its controller. The controller is the command center for the NAND memory, designating where each memory cell will read or write data and communicating with the interface that connects to the computer. Because NAND itself is imperfect as a media, how the controller responds and works to correct errors is a critical part of the design. Additionally, NAND has a finite number of writes per cell that can be made before the cell wears out. A well-designed controller will incorporate one of several types of wear-leveling, a technique that uses algorithms to manage the cells usage and spread the data throughout the NAND to maximize the SSDs lifespan.
Lastly, the electronics and ASIC assembly are the final area of components that put the SSD design together, integrating the NAND and its controller. The ASIC and electronics design will supply the necessary power/voltages to the SSD. Designs can vary here, as well as the quality of components selected. A poor ASIC design can lead to premature component failure.
MLC and SLC in the Enterprise
In early SSD design development, the positioning of SLC-based designs was meant for the highest performance enterprise applications, but remained as niche applications where storage cost itself was not a primary concern.
Most enterprises have system cost and IT budget restraints, and because of this, MLC-based solutions could potentially be a better fit. However, write endurance and reliability remain concerns with MLC in the enterprise. SSD designers recognize that if the issues around MLC could be resolved, that MLC devices would be ideally suited for the majority of storage systems at the upper Tier 0 transactional performance segments.
Enterprise vs. Client Class SSDs
For traditional hard disk drive (HDD) storage, reliability for enterprise-class designs is tested at full duty cycles, 24x7. Enterprise-class is also based on multi-drive environments, with an emphasis on random access patterns and mixed workloads. Non-enterprise, or client, class of storage is tested based on typical 8-hour-per-day use cycles. In addition, client-class storage is focused on single-drive environments.
SSD storage requirements are the same as HDDs in the enterprise, but standards have not existed until recently. This was a problem in the market because many SSDs were touted as being enterprise-class, although many did not function as true enterprise SSDs and experienced high failure rates during OEM qualifications or out in the field.
In September 2010, the JEDEC Solid State Association published two sets of standards for SSD endurance and reliability. JEDEC JESD218 and JESD219 address the standards needed to distinguish between SSD endurance in both enterprise and client application classes. Both standards documents are available for free download.
These standards define specific requirements for each application class, describe a test methodology, and create an SSD Endurance Rating that provides a standard comparison for SSD endurance based on application class.
By actively participating in the development of these standards and delivering a test path, OEM storage vendors and end users will be better served by finally having SSDs that can be tested, validated, and then confidently placed into the most appropriate storage environments.
Teresa Worth is a senior product marketing manager at Seagate, and David Szabados is a senior corporate communications manager at Seagate.