Flash Storage​

WINNING THE PERFORMANCE TRIFECTA: See how Hitachi VSP 5000 Series and Brocade Gen6 Fibre Channel accelerate NVMe performance for enterprise consumers
 

By Neil Salamack, Senior Product Marketing Manager, Hitachi Vantara

Marcus Thordal, Principal Solution Architect at Brocade

NEIL:
I’m joined by Marcus Thordal to talk about what’s happening in the flash storage space and how our joint enterprise customers gain real value from better understanding NVMe and why it matters. It is a great time to be in storage or memory or networking for that matter. We’re perhaps at a watershed moment, when more than half of all enterprise storage arrays are all-flash and crave continually greater benchmarks for performance—and along come these new capabilities for radically improved performance, scale and cost efficiencies. So, let’s hop right in, Marcus. We can start by introducing NVMe and the issues it can solve.

MARCUS:
Thanks Neil. Non-volatile memory express (NVMe) is an open portfolio of standards for transporting signals over storage, specifically non-volatile memory storage, such as SSDs, and is meant to streamline digital communications. Because NVMe is all about performance, its open protocol will supersede those of SCSI (SAS and SATA) for flash media and be consistently many times faster, with far less latency and power consumption per device than SAS.

Mainstream benchmarks right now show all-flash arrays (AFAs) with NVMe having 2-to-3x better IOPS,  2-to-2.5x greater bandwidth capacity, and up to 6x less latency than SAS. The latency advantage alone is something all users can benefit from, and in the right storage architecture, enterprises will gain extensive NVMe performance scale and consolidation opportunities.

As organizations prepare for the future’s unyielding digital pace, NVMe will be well-suited to harness next-gen storage class memory (SCM), which promises to outstrip today’s flash drive speeds. So, for industries steeped in high-performance computing, real-time analytics or large-scale transactional processing, NVMe warrants some real consideration for rapidly extracting intelligence from data and delivering value back to the business.

NEIL:
Digital transfer acceleration is demanded by everything from Wall Street to the Main Streets of Smart Cities. So many business-critical use cases come to mind. NVMe storage will feasibly be the preferred solution for:

·         AI and ML applications, and IoT deployments requiring higher read IOPS to support more queries in less time.

·         Financial institutions and online retailers needing to shave completion times for write-intensive, millisecond transactions faster than competitors.

·         Media and entertainment outlets demanding faster rendering and streaming.

·         Activities like oil exploration and extraction, or fraud detection and data mining needing to navigate expertly through millions of terabytes of potentially valuable raw data.

·         Machinery and manufacturing calculating millions of structural and computational fluid dynamics while eliminating costly errors.

·         Virtual machine farms wanting higher densities per server for better cost structures.

·         Consolidation of any workload on hypervisors for improved performance.

MARCUS:
And fields such as medical, pharmaceuticals and science have an unquenchable thirst for performance excellence to solve important problems. NVMe-based flash environments can support HPC’s computationally intense workloads and help to speed machine-learning tasks and automation simulations for things like:

·         Optimizing applications that allow researchers to simulate long-range climate modeling faster or search for new answers.

·         Using machine learning to expedite drug research and clinical trials for lifesaving medications

·         Personalizing emergency-room healthcare with readily available patient records.

NEIL:
The Virtual Storage Platform (VSP) 5000 series really accentuates what’s possible with NVMe. By now, our readers have heard about the VSP 5000 series being the fastest NVMe flash storage array on the planet, and I think it pays out on the performance trifecta of latency, throughput and IOPS. It screams with latency as low as 70 microseconds and uses Hitachi Accelerated Fabric high-speed interconnect to offload and speed IO communications between nodes in conjunction with the latest SVOS RF (v9) to manage IO across the fabric and allow performance to linearly scale as nodes get added.

Combine all that with 21 million IOPS and 69 petabytes of in-the-box capacity (with up to 279PBs for virtualized data to extend the life of older systems[i])! While a large margin of users might not need that many millions of IOPS, the lightning-fast response times will stay consistent even as capacity fills. We need to equip our customers for the new generation order-of-magnitude data demands. It’s what Hitachi envisioned when we designed the new VSP 5000 series’ scale-out digital architecture that orchestrates and accelerates any workload.

And it’s why we collaborated with Broadcom to help enterprise consumers capitalize on an NVMe flash-optimized, future-ready whole solution. Flash and NVMe are substantially faster in bandwidth, IOPS and latency than hard disk drives, which can overrun capacity of traditional network connections.

MARCUS:
Our unique NVMe alignment of the VSP 5000 series with Brocade Gen6 Fibre Channel really shines in this space. AFA storage workloads will continue to be performance hungry and today’s 10GbE and 16Gbps Fibre Channel won’t satisfy next-gen storage. Gen6 already deploys 32Gbps Fibre Channel networking and recent market testing comparing products using Gen6 FC with NVMe versus SCSI showed 40-60% higher IOPS and 30-40% lower latency[ii]. 

NEIL:
Hitachi offers a seamless upgrade path to NVMe over Fabric, so this might be a good time to explain NVMe backend and frontend, and the choices for fabric deployments.  NVMe backend is the interface for data transport between the storage controllers and the SSDs. It’s a faster alternative to SAS. By NVMe frontend we mean NVMe over a storage network like Fibre Channel or Ethernet.

For more information, please click here.

MARCUS:
As you know, Neil, Fibre Channel was built for reliable, high-speed delivery of bursty storage traffic whereas Ethernet was built as a broad platform for transporting many types of traffic and lacks native tools to optimize for the complexities of storage services. Nor do most data centers have dedicated Ethernet networks so it may require deployment of a net-new one.

One key point here is that only Fibre Channel allows NVMe and SCSI to run concurrently on the same network interface ports for a gradual migration and no disruption. So, for our enterprise customers already transporting their most valuable workloads via Brocade Fibre Channel, there is no need for a SAN rip-and-replace in order to adopt NVMe.

NEIL:  
Our integrated technologies promote an ideal way to embrace these new interfaces and truly deliver on the promise of future-ready performance. Honestly, there is so much more we can talk about in this NVMe space, especially around cost optimization of the VSP 5000 models and the Brocade Gen6 Fibre Channel switch.

I hope readers will join us next time as we cover the key supporting AI-based management software we use to accelerate just about everything in the data center.