Despite reaching state of the art status, the metrics for the best SSDs are still substantially behind the native performance of NAND Flash silicon. This is a direct result of SSD architectures which have from the outset been focused on complete backwards compatibility. While essential for initial technology adoption, the only way to achieve this backwards compatibility has been to emulate HDDs, which is also the architectural reason for the significant performance delta between any SSD and its Flash components.
Utilizing NAND Flash memory in storage devices requires performing processes such as garbage collection, wear leveling, and error management as well as handling the unique vendor and geometry specific attributes of different NAND silicon. Until now, these Flash management processes have been abstracted in a FTL, usually found embedded on the SSD, or in some instances found in host-side implementations, known as ‘host-based FTLs’.
FTLs perform these essential Flash management processes, but do so in an implementation that is redundant and overlaps with the host file system. This results in a dislocation that produces costly inefficiencies including poor latency, determinism (unpredictable latency spikes), IOPS, and bandwidth, wearing the Flash media out faster and reducing its usable life (endurance). FTLs also utilize overprovisioning to mitigate these inefficiencies, meaning that they require additional raw Flash capacity to perform these processes. These factors combine to unnecessarily increase both the acquisition cost, and TCO, of Flash storage.
However, FTL based SSDs have and will continue to play a critical role for Flash storage devices. By abstracting Flash management processes from the host system, a FTL-based SSD can be readily utilized by any host system without software modifications. This backwards compatibility will likely continue to be necessary for many applications where performance, cost and efficiency are secondary to legacy ‘plug and play’ requirements.
But for data center applications, where optimal performance, QoS, and cost are the key drivers, FTLs create a barrier that prevents ever realizing these objectives. Radian’s Symphonic Cooperative Flash Management (CFM) takes a different architectural approach that, while minimizing host-side modifications, overcomes the limitations of FTLs to deliver near native NAND Flash performance with the RAS and durability features required of a data center product. The result is a new Flash storage tier that can extend Flash into the future and throughout Next Gen data centers.
