Quad-Level Cell (QLC) technology represents the latest innovation in NAND flash storage. QLC is capable of storing 4 bits per cell, resulting in cheaper, denser storage than TLC, MLC, or SLC NAND alternatives. The cost and performance characteristics of QLC position it in between higher performance solid state devices and spinning disks. QLC flash has slightly slower performance than other flash technologies, but provides a significant boost over HDDs, while the density allows it to be a cost-effective solution for large capacities. Due to this, much has been made of QLC SSDs as a potential replacement for HDDs, and the eventual “death of disk”.
Currently, QLC has seen adoption by a few storage venders. NetApp’s latest addition to its FAS lineup, the FAS500f, ditches HDDs for QLC SSDs to provide a middle ground between its traditional hybrid FAS arrays, and the All Flash FAS (AFF) series. Similarly, Pure Storage released a capacity optimized version of its FlashArray, named FlashArray //C, which utilizes QLC NAND to provide a higher capacity, lower performance alternative to its //X models. An even more dedicated approach to QLC comes from storage startup VAST Data, which architected its entire Universal Storage system around using the technology as a backing store in combination with a high-performance persistent memory cache.
The arguments driving the use of QLC, primarily cost effectiveness and SSD performance, make sense and the sampling of adoption seen from various vendors serves as proof, however, implementing QLC is not as simple as a drop-in replacement for HDDs. Along with increased density and decreased performance, there is another key metric that changes with every level added to NAND flash: endurance.
While SSDs may be considered “more reliable” than HDDs due to the lack of moving parts, flash storage breaks down eventually as data is written and cleared from its cells. SLC NAND can withstand around 100,000 program/erase cycles, providing adequate endurance for most applications, but as more bits are stored per cell, the endurance of flash storage is decreased. MLC can withstand around 30,000 cycles, TLC can withstand around 5,000 cycles, and with QLC the endurance drops to around 1,000 cycles.
While the low number of program/erase cycles that QLC can withstand may be worrying, perhaps a more meaningful metric to consider is the drive writes per day (DWPD). The DWPD for QLC NAND is highly dependent on the type of write I/O. Small random I/O results in a lower DWPD, while larger sequential I/O results in a higher DWPD. Compared to the drive reads or writes per day (DRWPD) of HDDs, a similar metric to measure the durability of spinning disks, depending on the workload QLC devices can provide greater endurance.
The low endurance of QLC devices poses a clear challenge to implementing the technology in a useful, long lasting manner, however the DWPD measurements show that in some cases the issue can be overcome. When designing a QLC based storage system, careful planning must go into the system in order for the technology to be used effectively. New QLC storage systems must be designed from the ground up with the endurance issues in mind and existing systems which are repurposed for QLC must receive significant changes. Adjustments need to be made to the flash translation layer to reduce write amplification and provide proper wear leveling. Large, sequential writes cause less harm to flash devices, and therefore are likely necessary to compensate for QLC’s low endurance. One method of achieving this is to use a write cache, consisting of devices with greater endurance such as SLC NAND or persistent memory, to initially store and accumulate data until it can later be written sequentially to QLC.
Despite the challenges, QLC devices will likely see increased adoption by storage vendors as QLC NAND is developed with greater numbers of layers, leading to lower costs. As a number of QLC storage system offerings become available, it will become important for IT organizations to understand which systems efficiently implement the technology, along with the use cases in which QLC is most appropriate. QLC is best suited for large scale capacity needs and will likely be widely used in object storage systems looking to provide a performance increase over HDDs. It is the combination of needing cost-effective scale and increased performance that will make QLC the most useful. In the most high-performance systems, other types of NAND flash will be more effective. For use cases where scale and cost are much more pressing considerations than performance, HDDs will still remain competitive.
Even when used in ideal use cases, often in which the workload is more read intensive than write intensive, QLC’s endurance issues should not be ignored. IT organizations set to use QLC based systems must understand how the system is adjusting for the reduced durability of the devices. It will also become increasingly important to understand what warrantees and guarantees vendors are offering for these low endurance devices.
While the endurance issues of QLC appear daunting, a successful QLC implementation – one that accommodates the low endurance of the devices – can provide a significant upside with expected greater longevity than electro-mechanical devices. QLC provides cost effective, dense storage devices with greater performance than HDDs and has the potential to be a successful replacement for many disk-based use cases. Despite this, and what plenty of QLC marketing strategies have claimed, disk isn’t dead. HDDs are widely used in many storage systems and the technology will continue to see use as these systems are maintained and updated. Meanwhile, the endurance issues of QLC create a barrier for adoption that requires a well-planned architecture and may delay product delivery by many vendors. The translation layer adjustments needed to make QLC reliable can, and will, be done by a number of vendors looking to leverage this new technology, however, HDDs are not likely to be disappearing entirely anytime soon.
