IBM has announced that it is developing Non-Volatile Memory Express (NVMe) solutions to provide significantly lower latency storage.
NVMe is an interface protocol designed to replace the established SAS and SATA interfaces that are currently used for hard drives and SSDs. Coupled with the PCIe hardware backplane, NVMe uses parallelism and high queue depths to significantly reduce delays caused by data bottlenecks and move higher volumes of data within existing flash storage systems.
IBM has set itself to the task of optimizing the entire storage hierarchy, from the applications software to flash storage hardware, and is re-tooling the end-to-end storage stack to support NVMe. The company recognized years ago that both hardware and software would need to be redesigned to satisfy the needs of ultra-low latency data processing.
The company last year released products with this kind of optimization with the IBM FlashSystem A9000, which already includes features intrinsic to NVMe functionality.
IBM explains that the company is developing solutions with NVMe across its storage portfolio, which it plans to bring to market in the first half of 2018. By doing this, IBM tells us that it is leveraging its storage software leadership to ignite an industrywide leap in system performance.
What does this mean to you? Expect to see a massive conversion of storage to the NVMe interface over the next two years.
This will come about for two reasons: NVMe is extraordinarily fast compared to its SAS and SATA counterparts, and its price is coming in line with these two predecessors.
Although NVMe is a new protocol it evolved from the need to provide solid state storage with a better interface than was available using the HDD-oriented SAS and SATA protocols. Both RAID cards and some early SSDs took advantage of the PCI bus, which was originally developed to support the high-bandwidth needs of graphics cards. Since this was a new approach each card used its own unique communication protocol. Intel worked to develop a standardized protocol by forming the NVMe workgroup (of which IBM is a member) and that workgroup released the first version of the NVMe standard in 2011. Over time this standardization will lead to NVMe SSDs that are cost-competitive with SAS and SATA SSDs. When this happens IBM’s NVMe solutions should be in the right place at the right time to take advantage of this speed benefit.
The NVMe protocol has also been designed to support newer nonvolatile memory technologies that will be coming to the computing environment in the near future, making the standard even more imperative to support for tomorrow’s high-sped system solutions.
NVMe is the best way to communicate with flash today, and will be the preferred communication protocol for high performance solid state disks for the foreseeable future.
IBM is using this announcement to tell us that it is taking the bull by the horns. The company has committed itself not only to the use of this faster storage protocol, but it is also focusing on reconfiguring its software to get the most out of the use of NVMe. Existing software has been built around the assumption that storage is the slowest part of the system, so storage commands need not be optimized for speed. SSDs and future storage class memories turn this thinking on its head – now the storage stack has drawn the focus of storage developers since it could prevent solid state storage from delivering more than a fraction of its performance.
IBM is indeed a leader in the world of storage, and its moves will be mirrored by many. The SSD Guy expects to see a good number of upcoming announcements from the company over the next several months as it tunes all levels of its systems to take advantage of every bit of speed that solid-state storage has to offer.