Factors Driving Enterprise NVMeTM Growth Sponsored by NVM Express™ organization, the owner of NVMe™, NVMe-oF™ and NVMe-MI™ standards

Speaker Gary Kotzur

Overview With the increased availability and awareness of NVMeTM products, more enterprises are making the transition to NVMe. Join our panel discussion to learn how server and storage vendors are helping make this transition possible.

NVMeTM SSD Technology Trends Lower costs Natural market decreases Technology advances Capacity is increasing More layers More levels: TLC  QLC Die per package increasing Performance is increasing Media: SCM Interface: PCIe ® 3.0  4.0 transition, 5.0 is coming Power levels are sustaining Media Interfaces

Addressing Customer Expectation Gaps Enhanced Serviceability PCI-SIG® work group added new features All OS vendors have support plans Decreased Costs Infrastructure costs are dropping Switched and direct connectivity Modality of device ports More “Data Protection” Options Security Management

Expanding Ubiquity Wide adoption across all segments: Enterprise  Client NVMeTM support is expanding throughout Enterprise / Data Center products Native OS and OSV driver support Increased “Data Protection” options Native Fabric support  NVMe-oFTM Increased Media choices  Flash & SCM In summary  SSDs are converging on NVMe

Consolidation SSD Form Factor Incumbent Form Factor adoption: SD, XD Storage Device Interface NVMe-oFTM variants Modality of device ports

Challenges  Opportunities Flash Performance Density PCIe ® Distance Computational Storage Block  Key Value

Speaker Steve McQuerry

NVMeTM in Enterprise Storage Arrays Overview More storage Arrays are being built as all flash platforms for enterprise deployments These platforms are optimized for performance and capacity NVMeTM has become a key component when using flash in enterprise products Many all flash arrays are adopting NVMe as the preferred protocol NVMe-oFTM connectivity is providing additional benefits of extending the NVMe protocol outside of chassis directly to the client or to expansion shelves

NVMeTM Implementation Case Study We are looking at Pure Storage as a specific case study in the implementation and use of NVMeTM technologies. This is not a product endorsement by NVMe Express. As all flash became a component in block storage arrays, the initial protocol was still SCSI NVMe provided significant performance improvements over the SCSI protocol In order to get the benefits all flash arrays would need to transition users from an end to end SCSI experience to an end to end NVME experience

NVMeTM Transports NVMeTM Transports Memory Message Message & Memory Data & Commands/Responses use shared memory Commands/Responses use Capsules Data may use Capsules or Messages Commands/Responses use Capsules Data may use Capsules or Shared Memory Example: Examples: Examples: PCI Express Fibre Channel, TCP RDMA (InfiniBand, RoCE, iWARP) Fabric Message-Based Transports Capsule = Encapsulated NVMe Command/Completion within a transport message Data = Transport data exchange mechanism (if any)

NVMeTM Implementation Case Study The first step to providing NVMeTM capabilities within the Array Chassis From a buy or build perspective NVMe platform availability was limited until recently Customers were able to get the benefits of NVMe within the array, but the initiator, fabric, and in some instances expansion shelves were still using SCSI Controller CPU PCIe NVMe FLASH Array Chassis

NVMeTM over Fabrics Backend Implementation example The next step was to extend NVMeTM capabilities to the shelves Transport choices included PCIe or NVMe-oFTM, in the case of Pure Storage we chose NVMe-oF specifically NVMe™/RoCE This extended the NVMe experience and allowed extra capacity without sacrificing the protocol benefits NVMe-oF NVMe-oF Controller CPU Shelf CPU PCIe PCIe NVMe NVMe FLASH FLASH Finally, we are introducing the Direct Flash Shelf, a companion to the FlashArray //XR2, which enables us to extend the benefits of our Direct Flash Modules to capacity points beyond what can fit inside the base array. The Direct Flash Shelf employs a high-performance 50GbE RoCE interconnect which we use as a base for a fully functional NVMe-oF implementation to allow us to expand our capacity by 1 or 2 shelves, each of which can hold 28 direct flash modules. The direct flash shelf was a first for us in a few ways.  One is that it allowed us to develop an in-house, full NVMe-oF impelementation where we control both ends of the wire. That allowed us to focus on design, implementation, and performance in a controlled environment, and deliver a Pure-quality solution which allows our customers to reap the performance and capacity benefits of new technologies. In addition, we had to develop a new control plane to provide our expected  level of integration, simplicity, and ease of management for the shelf.                                                                                  Our next step will be to provide those same benefits directly when we introduce our NVMe-oF front end implementation, which we are working on now and expect to launch later this year. Array Chassis Array Shelf

NVMeTM over Fabrics Frontend Implementation example The finial step was to provide an end-to-end NVMeTM experience by exposing the volumes in the NVMe namespaces across the fabric This provides the host the added benefits of performance and efficiency of NVMe HOST CPU NVMe-oF NVMe-oF NVMe-oF Controller CPU Shelf CPU PCIe PCIe NVMe NVMe FLASH FLASH Array Chassis Array Shelf

Summary When working with an All Flash Array, it makes sense to leverage NVMeTM Customers are able to absorb NVMe into their environment without making significant changes to their infrastructure NVMe-oFTM makes it possible to extend the benefits of NVMe outside the chassis and all the way to the host to provide and end-to-end NVMe experience There is still work to be done on the OS Ecosystem and feature parity for NVMe-oF but the level of customer interest is increasing

Questions?