Solid State Devices (SSDs) are being rapidly adopted as a new tier of storage with 1/4th-1/10th the cost of a DRAM, and 50-100X better performance that HDDs. The technology landscape is undergoing a rapid transformation leading to reducing $/GB and $/IOPS. NAND-based Multi-Level Cells (MLC) are being increasingly used instead of Single Level Cells (SLCs) – MLCs have lower costs but reduced performance and life endurance. Storage vendors are compensating the reduced performance and reliability by implementing algorithms and ASICs for better wear-leveling, RAID techniques, T10 DIF/DIX support for data integrity.
At the driver interface level, standards such as NVMe are pushing for standardization of block-based drivers, which further reduce the software stack latencies compared to traditional SCSI. Another optimization is in the placement of the Flash Translation Layer (FTL) within the host driver (the Fusion-io model) instead of the physical Flash device.
SSDs are creating a new set of use-cases: Host-based caching; SSD as Tier 0 storage; All Flash Storage arrays; Specialized appliances/gateways are a few popular examples. SSDs are creating opportunities to innovate across the stack: From changing application APIs for low latency storage, to workload specific optimizations such as caching the "golden image" in Virtual Desktop Environments, to optimizations of the core IO stack and drivers within the host or hypervisor.
Decreasing price point (w/ MLCs), increasing densities, and growing adoption in enterprises ensures interesting innovation across all layers of the IO stack (including the application-tier) in the coming months!
No comments:
Post a Comment