Top 5 Reasons to Consider NVM Express over Fabrics For Your Cloud Data Center White Paper
Top 5 Reasons to Consider NVM Express over Fabrics For Your Cloud Data Center Major transformations are occurring in cloud storage architectures. NVM Express (NVMe ) SSD adoption in the cloud is growing quickly due to its high performance enabled by a light-weight, standards-based interface. The NVMe over Fabrics (NVMeoF ) protocol extends those values to the network, making shared accelerated storage a new option for traditionally DAS-based cloud architectures. The move from DAS-based storage architecture to a disaggregated, shared-storage model presents several benefits. Demand for responsive cloud-based applications and the rapid migration to microservices and container-based applications are taxing existing storage architectures to deliver high performance, low latency, reliability and scalability. For the past several years, this has resulted in a dramatic transition toward solid-state drives (SSDs). Now, that move to SSDs has to accelerate further. SSDs based on SAS or SATA interfaces need a turbo boost in order to fulfill nextgeneration workload requirements, especially the ability of those interfaces to handle concurrent IO to a single device. 1 Enter NVMe, which speeds performance by using the high-throughput PCI Express (PCIe ) bus rather than a dedicated storage bus that requires protocol translation. The resulting faster bandwidth and lower latency make NVMe-based SSDs great solutions for such use cases and workloads as analytics, high-performance computing, databases and containerized infrastructures. NVMe-oF with disaggregated storage enables cloud data centers to fully utilize NVMe flash storage. This means that interest in and demand for NVMe-based SSDs is surging. IDC predicts that NVMe will become the mainstream foundation technology for enterprise storage by 2020 because more and more workloads will require the high throughput, low latency and resilience of NVMe. As a result, IDC projects that NVMe-based SSDs will become a nearly $10 billion market by 2020. 2 2/6
With the NVMe-oF protocol, the performance and standards-based benefits of NVMe have been extended to the network. For cloud infrastructure decision makers, this represents a great opportunity to realize the benefits of shared storage without sacrificing the performance and cost benefits of DAS. The NVMe protocol has been nearly 10 years in the making and has wide acceptance and support across the storage vendor industry. The NVMe-oF specification is developed and supported through the NVMe Fabrics Working Group. This organization has also published a roadmap to give cloud infrastructure teams a clear vision into how to plan for the upcoming hardware and software enhancements to NVMe-oF. Whether your organization is upgrading your cloud deployments or building new ones, you should seriously consider deploying a NVMe-oF Shared Accelerated Storage solution. Here are five reasons why: 1. Realize the benefits of disaggregation. Direct attached storage (DAS) has been the most popular storage architecture for cloud deployments around the world due to its low cost, high performance and simplicity. A major trade-off for benefits of DAS are that CPU and storage are essentially locked together for the life of their deployment. The data storage on DAS can be moved to another server, but in the world of terabytes, it isn t a practical solution. Cloud infrastructure teams have to take extraordinary steps to overcome the limitations of DAS. This means underutilized SSDs in the server or underutilized CPUs that need more storage performance or capacity. The most significant of these is overprovisioning the entire data center infrastructure for peak workloads. The solution requires the SSDs to be pooled and shared across the any number of servers in the cloud infrastructure. However, to date that disaggregated solution has not been practical. Disaggregation is facilitated by the wide spread adoption of 25 Gigabit Ethernet (GbE) and higher network infrastructure, the introduction of the NVMe-oF protocol and the wide adoption of high performance, high capacity NVMe SSDs. This carries a key benefit: less need to overprovision storage to meet ever changing capacity and performance requirements. Disaggregated storage is far easier to deploy, manage and scale than legacy DAS architectures, and helps to reduce and even eliminate stranded storage and underutilized CPUs. This efficient allocation of high-performance assets helps to control CapEx costs and makes it easier to pay as you go when looking to expand data center storage capacity. Most critically, the shared accelerated storage architecture allows the introduction of NVMe as a service in the cloud infrastructure, enabling applications to scale up and down independently. This allows a dramatic increase in CPU and SSD utilization, eliminating the need to overbuild for peak demand. 3/6
2. Key applications and use cases will benefit from improved storage performance. The introduction of SSDs into data center storage systems helped massive scaling of essential workloads such as cloud databases, analytics and machine learning. These and other workloads demand even better performance, higher reliability and NVMe-oF improves economic efficiency lower latency. This demand plus the ability to support a wide range of and provides better utilization of compute network types without compromises on performance or management and storage resources in the data center. makes NVMe-oF a great solution for NoSQL databases, containerized/ virtualized applications and machine learning. It also aligns closely with unique applications and use cases in vertical markets rapidly adopting cloud architectures such as cloud services, web services, financial services, telecommunications and massively multiplayer online (MMO) gaming. NVMe-based SSDs offer the unique combination of benefits in performance, latency and reliability essential to these and other workloads. Plus, NVMe-oF enables this performance to be realized in a network environment, with less than 20µs incremental latency (note: typical SATA SSD has 200µs latency). 3. Superior economics. For years, storage architects worked to reduce the high cost of data at rest in order to improve total cost of ownership for storage. NVMe-oF helps extend that practice to cloud architectures. The return on investment is to get value out of the data at the lowest cost. The utilization of shared pools of SSD storage reduces CapEx costs, while also reducing operating expenses because of associated smaller data center footprint, lower power consumption and cooling costs. NVMe-oF also enables faster, easier and more efficient expansion, as well as automated orchestration tools that free-up in-house storage management staff to handle other responsibilities. New software functionality enabled by NVMe-oF can also extend SSD drive life, further reducing CapEx and sustaining performance, even in hyperscale workloads over time. The result is impressive: lower TCO, faster ROI, reduced CapEx and OpEx. NVMeoF improves economic efficiency and provides better utilization of compute and storage resources in the data center. Because of cost-savings, you can use more flash for your cloud applications, enabling greater data velocity and better scaling for peak demands. 4/6
4. Performance and QoS upgrades. Cloud workloads demand more from storage and require consistent, sustained performance. That s where NVMebased SSDs and NVMe-oF shine; the higher IOPS, faster throughput and lower latency are hallmarks of NVMe, and the NVMe-oF architecture is designed to keep latency low across the network. Championed by industry leaders including Facebook and Toshiba, IO Determinism, a Quality of Service (QoS) innovation developed by the base NVMe technical workgroup, will continue to lower latency and improve consistency in scale-out cloud deployments. 5. Achieve agility and scalability of public cloud with private cloud deployment. Updating your cloud storage architecture can satisfy your customers demand for instant scalability like the public cloud and your manager s requirement for budget control, security and compliance in the private cloud. NVMe-oF provides many attractive benefits, including greater container density resulting in more peak capacity from the same CapEx. New software features enable dynamic container node orchestration for stateful applications with persistent high performance storage. NVMe-oF allows for dynamic management and CPU and storage allocation based on users actual demand. Plus, NVMe-oF enables seamless integration with automation frameworks, like Kubernetes and OpenStack, allowing for hands-free operation which users expect from their cloud applications. Toshiba s leadership role in NVMe over Fabrics For decades, Toshiba has been a global leader in the development of storage technology, including helping to pioneer the development of flash memory technology and NVMe-based SSDs. Toshiba is a promoter company and a member of the board of NVM Express, Inc., which is spearheading the development and distribution of technical standards for NVMe for storage. Toshiba has extended its technical contributions from hardware to software with a series of important announcements regarding NVMe-oF. In mid-2017, Toshiba introduced software technology designed to facilitate deployment of NVMe-oF for cloud data centers. As the inventor of flash memory in 1984, Toshiba draws upon a history of substantial technical innovation in its broad portfolio of NVMe solutions. Toshiba provides write-intensive, read-intensive and mixed-use SSD drives across a wide range of applications and use cases, utilizing the SAS, SATA and NVMe interfaces. 5/6
Conclusion NVMe-based SSDs and NVMe-oF represent major transformations in cloud data centers. These technologies are gaining acceptance for their ability to help customers re-architect their data centers to meet the demanding needs of cloud workloads that require sustained high performance, low latency, extreme resilience and impressive cost efficiencies. As storage architects move from legacy DAS-based storage to a disaggregated model that facilitates shared storage across multiple network protocols and supporting multiple orchestration and provisioning frameworks, NVMe-oF increasingly will be the preferred approach. Data center decision makers should consider each of the options as starting points for their journey toward NVMe-based SSDs and NVMe-oF cloud deployments. Toshiba offers decades of experience in storage and is a pioneer in the development of NVMe-based SSDs and in the creation and implementation of standards for NVMe-oF architecture, hardware and software. To learn more about Toshiba s NVMe-based SSDs and its NVMe-oF software capabilities, please visit: KumoScale.com 1 How NVMe over Fabrics will change the storage environment, www.searchstorage.com, September 2017 2 NVMe in Enterprise Storage Systems, IDC KumoScale is a trademark of Toshiba Memory Corporation. NVM Express, NVMe and NVMe-oF are trademarks of NVM Express, Inc. PCIe and PCI Express are registered trademarks and/ or service marks of PCI-SIG. Kubernetes is a registered trademarks of Linus Torvalds. The OpenStack Word Mark is either registered trademark/service mark or trademark/service mark of the OpenStack Foundation, in the United States and other countries. We are not affiliated with, endorsed or sponsored by the OpenStack Foundation, or the OpenStack community. All other company names, product names, and service names mentioned herein may be trademarks of their respective companies. Information in this document including products, availability, specifications, technical/application data and contacts are current and believed accurate on the date of publication, but is subject to change without prior notice. 6/6