Post-Trade Processing Modernization – Fast tracking the benefits of Analytics & AI while driving huge operational efficiencies Is there a financial services organization in the world who...
NVMe is not only poised to unlock the next generation of performance and density, but also represents a technology that we believe will have a profound impact on all-flash arrays in the coming years. With the NVMe transition starting in 2017, if you’re buying a storage array now that has no upgrade path to full NVMe, your investment is likely being thrown away and the benefits of NVMe foregone or deferred by several years. Ready to learn more about NVMe?
Today, we’re excited to share the second episode of Outside The Box – our innovation video series for techies by techies – with a deep-dive on NVMe. You’ll hear from experts on NVMe, including our founder John (Coz) Colgrove and Howard Marks from DeepStorage. Watch here to get up to speed on NVMe, or read on.
Current storage systems use SAS (serial-attached SCSI) links from their controller processors to communicate to flash. SAS uses the legacy protocol SCSI, which was designed for disk, yet presents a serial bottleneck for communicating to flash. Each connection from CPU core to flash is limited by the SAS host bus adapter (HBA) and synchronized locking. This serial bottleneck creates performance bottlenecks within the back-end of all-flash arrays. The NVMe (over PCIe) protocol holds the promise to eliminate the SAS (SCSI) bottleneck. NVMe brings massive parallelism with up to 64K queues and lockless connections that can provide each CPU core with dedicated queue access to each SSD.
In addition to NVMe, NVMe/F (NVMe channeled over RDMA-enabled fast Ethernet, FC, Infiniband or other fabrics) is another new protocol for dramatically faster, SCSI-less server-to-storage connections. We expect that NVMe within storage arrays will materialize first, and NVMe/F for server-to-storage connection will lag a bit as the standard was later to develop and it has a dependency on multi-vendor interoperability. The massive parallelism unlocked by NVMe will be required to take advantage of future technological advances (super-dense SSDs, modern multi-core CPUs, new memory technologies and high-speed interconnects).
WHAT WILL NVMe BRING?
NVMe will bring many benefits to customers’ applications and data centers including:
WHEN IS THE SHIFT TO NVMe?
In Outside The Box episode about NVMe, Howard Marks makes the case that the tipping point to NVMe is coming soon, even as soon as 2018. Similarly, leading industry analysts forecast that NVMe will become the leading interface protocol for flash by 2019. We, of course, are investing to lead this transition in 2017! NVMe is already present in consumer devices such as laptops/desktops, hyper-scaler clouds are using and/or planning to use NVMe, and the enterprise will not be far behind.
A tested and proven product design philosophy is to follow the volume, as higher volume often means lower cost, more R&D investment, and faster innovation.
HOW WILL PURE DO NVMe?
Just like we did with flash in 2009, our approach to NVMe is different than other vendors, many of whom are downplaying the near-term benefits of NVMe or are approaching NVMe as an exotic, niche technology for the performance elite. We believe in NVMe for mainstream, not niche, exotic use cases. We also believe in NVMe without compromise – with 6 9’s reliability and Tier1 data services such as data reduction, snapshots, disaster recovery, etc.
At Pure, EvergreenTM Storage isn’t just a marketing slogan – it’s built upon an Evergreen architectural foundation and coupled with a unique business model that has been delivering a competitive advantage to our customers for years. Customers have been able to benefit non-disruptively from new technology innovations such as several compute/controller generations, flash density gains, MLC to TLC shift, etc. Similarly and as a natural extension of Evergreen Storage, we believe that NVMe should be a non-disruptive, Evergreen upgrade, not a forklift migration. This way, customers can protect their investment, and benefit from NVMe without a re-buy or risky, time-consuming data migration. You can learn more about our approach to NVMe here.
In anticipation of the now inevitable shift to NVMe, we engineered FlashArray//M to be NVMe-Ready from the beginning, starting more than three years ago:
We believe the process of retrofitting NVMe will be a difficult one for legacy storage arrays, likely even more difficult than the original transition to flash. Unlike in the flash transition, where SSD vendors shipped flash that simply emulated a hard disk drive, NVMe requires both a change in the “wire” for connecting SSDs, as well as a whole new protocol over that wire. This transition will be extremely difficult for legacy arrays to bridge, let alone make the leap to NVMe in a non-disruptive manner.
As a result of the foresight put into the //M architecture, customers will be able to upgrade any FlashArray//M to NVMe-enabled controllers and capacity without a forklift upgrade or disruptive migration. Pretty cool! In fact, we plan to deliver NVMe upgrades – NVMe controllers and NVMe Flash Modules – for FlashArray//M by end of 2017. Combined these upgrades provide the building blocks for a full NVMe //M!
Customers will be able to add NVMe to integrate NVMe and SAS into a mixed system or consolidate into a denser, NVMe system with few, simple steps. For example, take the case of an existing FlashArray//M with base chassis plus an expansion shelf with SAS Flash Modules. By adding an empty shelf, the SAS Flash Modules can be moved non-disruptively from the base chassis to the empty shelf (Purity Operating Environment already support this). Once the //M base chassis is empty, the //M controllers are upgraded to NVMe-enabled controllers (again the controller upgrade is a non-disruptive operation on the FlashArray and something that our customers absolutely love about the product). Now, NVMe Flash Modules are added to the //M base chassis, and a mixed NVMe and SAS //M is born. Customers can choose to go one step further and take advantage of Capacity Consolidation* to trade-in older SAS flash and consolidate into denser NVMe, and thereby complete the journey to an NVMe //M – all performed non-disruptively and with investment protection!
We believe that anyone buying a storage array in 2016 needs to ensure that it can be upgraded to NVMe in 2017 or beyond, in order to protect their investment. Pure stands by our products and our customers so we are guaranteeing that every new FlashArray//M will be upgradable to NVMe in two ways: 1) controller upgrade to NVMe with eligible Forever Flash (Free Every Three) renewal, or 2) upgrade sooner to NVMe controller with Upgrade Flex bundles. One way or another, an investment in the FlashArray today means an upgrade to NVMe in the future. And, if we can’t make the upgrade work, we will give customers a replacement NVMe array – free!
As you can tell, we are incredibly excited about NVMe and we look forward to sharing more (a lot more) on NVMe in 2017! Engage us if you want to hear more from our experts on NVMe, or simply want to learn how the FlashArray can help you gain a competitive advantage – today and in the future.
Pure Storage, Evergreen and the “P” Logo mark are trademarks of Pure Storage, Inc. The NVM Express™ design mark and NVMe™ word mark are trademarks of NVM Express, Inc. Unreleased products, features, or functions referenced in this post, our website or other public statements that are not currently available are subject to change at Pure Storage’s discretion and may not be delivered as planned or at all. The availability and timing of NVMe upgrades for various capacity points or models may vary. Customers who purchase Pure Storage’s product offerings should make their purchase decisions based upon products, features and functions that are currently available, or upon the current Evergreen Storage Program Terms, which includes the NVMe-Ready Guarantee. *See Evergreen Storage Program Terms for details, including regarding NVMe-Ready Guarantee, Forever Flash and Capacity Consolidation.