The Future of Flash Storage
In a world that already has close to 5 ZB of data in existence and will arrive at 44 ZB within a decade, it is clear that to stay ahead of burgeoning storage demands, flash will have to get faster, cheaper, denser and have greater longevity.
“The combination of trends such as BYOD, the proliferation of mobile devices, the Internet of Things (IoT), and higher-resolution photography and video (to name a few), are leading to massive data growth for on premise storage as well as cloud-based storage,” said Steve Bohac, director, product management and strategy at Violin Memory.
Part of the problem is that the storage industry got a little lazy when solid state drives (SSD) burst onto the scene. Flash was packaged in SSDs that fit nicely into the existing form factor of hard disk drives (HDDs). They then slotted perfectly into the slots of storage arrays, providing a fine boost in performance with a minimum of effort.
“By definition, an SSD is a device that emulates disk behavior but uses memory (in this case, NAND flash) as the storage medium,” said Bohac. “By using SSDs, legacy and hybrid vendors have bargained away a key benefit of using flash, bottlenecking capacity and performance.”
He pointed out that a disk drive has one actuator arm, and it can thus only perform a read or a write operation at any one time. Invented around 60 years ago, the disk drive cannot read and write at the same time. That’s why many vendors are gravitating more to PCIe-based flash and similar configurations to take greater advantage of flash. For example, Violin has developed its Flash Fabric Architecture, which is the foundation of its Flash Storage Platform (FSP), which Bohac said is unrestrained and uncompromised by disk architectures.
But freeing itself from older form factors is far from the only road open to flash innovation. There is a lot going on with regard to increased density. Flash is formed from collections of memory cells that hold different amounts of data. Single-level cell (SLC) flash holds one bit per cell, but can deal with many read and write cycles before it wears out. Multi-level cell (MLC) flash holds more bits per cell and is cheaper than SLC, but wears out faster. There is also eMLC which seeks to combine the benefits of MLC and adds more durability. Triple-level cell (TLC), which has price and capacity advantages, has also been developed but at the expense of speed and durability. Storage vendors have been working on better controller technology to erase some of these drawbacks and improve the error correction of TLC. That’s why DRAMeXchange predicts it will comprise half of all flash production by the end of this year.
Beyond the type of cells, manufacturers have figured out ways to stack more layers of flash vertically. Intel and Micron Technology, for example, have unveiled 3D NAND which raises capacity by three times while curtailing power consumption.
“This enables gum stick-sized SSDs with more than 3.5TB of storage or standard 2.5-inch SSDs able to hold 10TB,” said Doug Rollins, SSD marketing engineer at Micron Technology.
Further, Intel and Micron are working on a new class of flash-like memory known as 3D XPoint which opens the door to data being written in smaller sizes, enabling faster and more efficient read/write processes. Result: a memory technology promises to be ten times denser and up to 1,000 times faster than conventional flash.
Such developments are straining the communication channels of storage devices. Created in an era when Internet users were happy with 28k dial-up connections, the communication protocols that transport data between devices were overdue for an overhaul. Accordingly, NVMe is beginning to hit the market as a way to bring these data transport mechanisms up to the velocity of modern processors and flash architectures. Early NVMe products have demonstrated up to six times greater random and sequential read/write performance.
“One of the advantages of the work being done with NVMe is the improvements in efficiency for addressing flash and other non-volatile memory,” said J. Metz, R&D engineer for storage at Cisco Systems. “Combine that with a dramatic reduction in power and you have a recipe for faster, less power-hungry devices.”
But this isn’t the end of the flash innovation line. The efforts of Korean researchers to lower the price of flash means that smartphones record information only when necessary and in much smaller batches, while extending the lifespan of flash storage by up to 39%. Called "Write Ahead Logging Direct IO" (WALDIO), the trick is to record information only when necessary, and in smaller batches. This could have major implications for flash storage, too, in the years ahead.
Photo courtesy of Shutterstock.