With memory on servers being used increasingly for primary storage, it was only a matter of time before the best attributes of DRAM and Flash memory were unified in a single product.
At the Flash Memory Summit, PMC announced today a series of PMC Flashtec NVRAM drives to provide server-side storage using DRAM, which is capable of supporting 10 million I/Os per second.
Roger Peene, senior director of marketing for PMC, says that by leveraging an NVMe interface for the PCIe 3.0 bus alongside PMC controllers, PMC Flashteh NVRAM drives deliver more than 10 million IOPS for small, random access transactions when used as persistent memory. When the PMC Flashtech NVRAM is used as an NVMe block device, it delivers one million 4KB IOPS. In both instances, Peene says those IOPs represent sustained rather than peak performance levels.
Peene says PMC expects that this new class of drives, which are based on a type of non-volatile random access memory (NVRAM), will be sold both through its OEM server partners and directly to IT organizations with enough of their own engineering talent to implement it. For example, Peene envisions scenarios where companies that build their own servers to support Web applications that run at scale would be able to incorporate PMC Flashtec NVRAM drives using their custom BIOS implementation. The use of NVRAM on those servers would require no changes to the operating systems that run on them because the PMC drives look like standard DRAM memory to an application.
Scheduled to be available in 4GB, 8GB and 16GB capacities, which will be in production by the end of the year, the PMC Flashtec NVRAM drives are another example of how in-memory computing is transforming the way data is processed and stored. Given the costs and current storage capacity limitations, NVRAM drives attached directly to servers and as an alternative to solid-state drives (SSDs), they will be used to process the most frequently accessed or “hot” data in the enterprise.
The challenge for a lot of organizations isn’t going to be embracing server-side memory as a storage medium, but rather figuring out where all the hot data that can best leverage that memory is currently being processed on their existing systems.