| |

Enterprise SSDs

 August 10, 2011, Flash Memory Summit, Santa Clara, CA—Scott Stetzer from STEC proposed storage, system, and network designers need to develop new ideas about when, where and how to use SSDs. Fitting SSD technology to the application layer can bring new value and reduce TCO.

Data centers are changing and now have to deal with IOPS two orders of magnitude greater than previous generations. SSDs are helping by enabling auto-tiering with reduced footprints and reduced costs. The system architecture must change to take advantage of these performance enhancements. One change is MLC devices are now moving into enterprise applications. This is a messy process because there are many performance, reliability, and endurance issues. These problems are reduced when a high-quality controller is combined with the flash into an SSD for increased functionality.

The industry is seeing a technology shift and should experience consolidation over time after this initial burst of companies and products. SSDs benefit from advances in process geometry, which makes it easier to increase capacities and reduce costs. The problem is fine pitch devices have more difficulties with reliability, necessitating the addition of more ECC, DSP, and other signal enhancement technologies.

There is a growing gap in performance and reliability between eMLC (TLC), MLC, and SLC devices and all need advanced management design into their controllers. These controllers will not only increase capacity, but improve reliability and endurance by changing from simple level counters to active cell management functions.

Big data is now a growing to the zetaB sizes, which is equivalent of 200 billion HD movies. These data will reside in various platforms, on servers, and in large storage arrays. The issues for the systems are capacity, energy consumption, performance, etc. These systems need new architectures that incorporate SSDs with auto tiering capabilities. Drive caches are moving into the servers to speed up data retrieval and reduce system IOPS.

SSDs are also moving into virtual machines with the justification being reduction in TCO. Cloud services will see changes in OS and databases to take advantages of the changes in storage and accelerate applications. The cost for solid-state storage is dropping. MOC devices currently cost about $10 per gigabyte and are expected to drop to $4 in the near future. TLC devices will drop this even further to $1.50 per gigabyte. These cost reductions will require significant increases in controller complexity to include statistical degradation models and active load management. Eventually, all storage in the enterprise will be some form of NAND flash, a mix of SLC and MLC for tier 0 and tier 1 functions and TLC for static data.

Similar Posts