SSDs in traditional storage systems, either as hybrid arrays or all flash arrays (AFAs) have been in the news for a number of years now, becoming mainstream solutions options to improve storage performance and total cost of ownership (TCO) for many IT datacenter workloads. However, the network latency for servers to access this data, combined with SSDs sitting behind bottlenecks such as SATA or SAS interfaces have not tapped into the full performance capabilities of flash.
One solution is direct-attached flash, where the individual servers can access this high speed storage directly, but this methodology also makes sharing the pool of storage either difficult or subject to similar network latencies of traditional networked storage. What is emerging now is a class of storage called “Rack-scale flash”, with notable new offerings from EMC’s DSSD division as well as AFA upstart Pure Storage, among others. Rack-scale flash promises to provide a shared pool of storage accessible by a number of servers over a high-speed connection either within a single rack or amongst neighboring racks. Both DSSD’s and Pure Storage’s offerings utilize custom designed flash modules, not standard SSDs, communicating either via NVMe or proprietary protocols over PCIe. Multiple modules combine together within a single shelf to provide high-speed sharable storage.
The EMC DSSD D5 connects with other servers through NVMe over Fabric (NVMeoF) for maximum performance, whereas the Pure Storage’s FlashBlade uses 40 Gb Ethernet, promising easy and affordable scale-out performance storage capabilities. Each solution targets different workloads and use cases, but the trend toward custom-designed flash modules may gain some traction over time and further diversify, and complicate, the forms of flash utilized by enterprise customers.