One of the most memorable lines from Star Wars: The Empire Strikes Back comes when Yoda challenges Luke to use the Force to lift his X-wing fighter out of the Dagobah swamp. Luke doesn't think it can be done and Yoda tells him, “You must unlearn what you have learned”. Luke needed to let go of his preconceived notions and assumptions about what was possible with the Force, in order use it effectively.
This is exactly where many customers find themselves when it comes to SSDs. They have heard (or learned from past experiences) about the limitations of NAND flash and have concerns about what how they can use it. However, SSD technology has advanced significantly, especially in the last few years. Advances in controller technology and SSD architecture have overcome many of the biggest challenges inherent to NAND flash. Many conceptions and ideas users have about what is important to look for in SSDs is now outdated. You must unlearn what you have learned.
Episode I: Endurance
One of the most challenging topics in SSDs is drive endurance. A few years ago, choosing an SSD was as simple as choosing between SLC and MLC NAND. Generally, SLC NAND was specified by NAND manufacturers with 100,000 program/erase (p/e) cycles while MLC NAND was only 10,000 p/e cycles – an order of magnitude difference. At the drive level, SLC-based SSDs were specified with 10x to 200x (up to 2M p/e cycles!) the endurance of MLC-based SSDs. In general, the mindset was that SLC NAND was necessary for most applications and MLC NAND was only suitable for consumer USB flash drives and SD cards for digital cameras. In reality, for most applications, few customers understood their system workloads well enough to determine how long an MLC-based SSD would last or if SLC endurance was really needed. System engineers just knew that SLC would give them a wide margin and peace of mind.
Today, innovations in NAND flash components, SSD controller technology, and SSD architectures have made it unnecessary for SSD users to consider the endurance in p/e cycles of the underlying NAND itself. NAND manufacturers like Toshiba have introduced NAND components of different grades like enterprise MLC (eMLC), or with built-in management like Toshiba SmartNAND™ -- devices which may have different endurance and data retention profiles depending on use case. Controller technologies such as page mapping, over-provisioning, DSP techniques, and more robust ECC such as Toshiba's QSBC™ (Quadruple Swing-By Coding), to name a few, extend the useable life of NAND. New SSD architectures may mix different grades of NAND flash for performance or drive life. With these innovations, the characteristics and specifications of the drive itself are far more important than the NAND.
The well documented performance benefits of SSDs have made the technology popular in enterprise applications. One effect of growing SSD adoption in the enterprise market has been a better understanding of workloads and usage models. These efforts to understand workloads were borne out of necessity to lower system cost by utilizing MLC-based drives. With this understanding, and with the aforementioned advances in SSD technology, it has become widely accepted that MLC-based SSDs are suitable for nearly all mainstream applications, as well as enterprise applications. SSD suppliers now target different application workloads with multiple classes of SSD, and SSD specifications now reflect this mindset. The major suppliers in the industry have migrated to specifying endurance in terms of the following:
- Endurance Rating, expressed in Terabytes Written (TBW) or Petabytes written (PBW), as defined by JEDEC JESD218 specifies “the maximum number of terabytes that may be written by a host to the SSD, using the workload specified for the application class” while still meeting the requirements of the application class.
- Drive Writes Per Day (DWPD), specifies the maximum amount of data that can be written by the host to the drive over a particular time period (5 years is typical), based on a given workload such as a JEDEC JESD219 defined workload, and expressed in terms of the drive's capacity.
Toshiba specifies its Enterprise SSDs with TBW and DWPD based on the JEDEC standards to help users to make more accurate comparisons between SSD products. Unfortunately, not every SSD manufacturer in the industry follows this practice.
What is the benefit of this focus on workload, TBW, and DWPD? Ultimately, it is about saving cost. As a system designer, understanding your workload enables you choose what type of SSD you really need, and how much SSD you really need, which can save you cost.
May the Force be with you, young Padawan.
Stay tuned for Episode II.
The views expressed on this post are my own and do not necessarily reflect the views of TAEC or Toshiba Corporation.