February 25, 2013 4:59 PM
Atoms vs Electrons: How flash memory is different from disk
The debate about flash versus disk has morphed significantly over the past few years. We have now shifted from a discussion on whether flash could even be used in the enterprise, to a conversation on price versus performance, to the acceptance that flash is now a crucial building block of the modern software defined data centre.
A good deal of this discussion has paralleled the reduction in the cost of NAND flash. As flash solutions became more cost effective, organisations saw the ROI along with the performance they expected from the investment and naturally started to make the shift. Let’s take a look at how architectures are delivering innovation in today’s datacentres.
Flash Memory Benefits
Flash memory offers a number of benefits over disk, from performance boosts to operating cost savings. Without mechanical parts, flash can deliver data orders of magnitude faster. When you no longer need racks of disks for performance, you don’t need as much space for storage in your data centre, so you can save more space for the CPUs that actually process your data. Also, since flash doesn’t spin like a disk, it doesn’t create friction, so flash generates a lot less heat than a spindle of disks. Less heat means less cooling, which means less budget spent on energy.
Today, there are many options for integrating flash into a data centre. For optimal performance it can be added to the server, it can be used as a cache to back end storage in both physical and virtual servers, and it can also be used as shared storage. Companies around the world are adding flash to their IT infrastructure in numerous combinations of these approaches.
The Science and Software Behind Flash
Part of the reason flash memory is so fast is that with NAND, we are moving electrons rather than atoms. Moving an electron takes less energy than moving a physical atom, so it can happen much faster with much less force. As NAND flash continues to become more affordable, disk will become an archival storage tier, similar to tape, as the lowest common denominator in the performance storage space. This is similar to when disk first entered the market and companies still used tape drives to warehouse data backups. Some still use tape today, and we would expect to see disk start to occupy that archival niche over time.
While it sounds simple to talk about moving electrons, the amount of electrons moving on a flash memory device is extremely large. With electrons representing each zero or one, it takes sophisticated software to ensure that the data stored on those electrons is safe and that the flash memory device is reliable. While some SSD vendors do basic adjustments to the SAS and SATA protocols used to integrate disks into storage systems, leaders focus heavily on developing systems that optimise the software that controls the flash memory platform and integrates it into the server. The result is impressive application acceleration that can make a significant difference to businesses in the information age.
Improving Reliability for the Datacentre
Another area of concern with flash memory is reliability. Just like hard disk drives, flash drives wear out over time. Disks can fail because of mechanical issues like head crashes, where the head makes contact with the rotating platter, or motor failure. In flash products, as electrons are moved, sometimes they stay behind and render a NAND cell useless. These errant NAND cells add up over time, but again using sophisticated software, NAND can be managed so that all cells wear out around the same time. This wear leveling technique means a drive can retain more space throughout its life, without causing decreases in performance, while also helping IT managers predict and plan around the life of their flash solutions. Essentially, flash failure is much predictable than disk failure.
The physical differences of flash versus disk, electrons versus atoms, solid state versus mechanical, are what set the two mediums apart. These differences are also why flash should not be limited by disk-era primitives and protocols. Rather, the potential of flash is optimised when it is integrated as an entirely new, persistent memory tier. As companies begin to develop new primitives and protocols that take advantage of flash memory’s capabilities, they will be able to create applications that before were unthinkable - just by moving electrons
Posted by Mat Young - Senior Director of Products at Fusion-io