Doubling Down on Our Unified Flash Fabric
by Dev Datta – Product Management

Nimble’s Unified Flash Fabric brings flash to all enterprise applications by unifying All Flash and Adaptive Flash (Hybrid) arrays into a single consolidation architecture with common data services. The Unified Flash Fabric was made possible earlier this year when we introduced an All Flash array that is a generation beyond anything in the market. Today, we are doubling down on the Unified Flash Fabric with the introduction of a new generation of Adaptive Flash arrays (the focus of this blog) and a new entry-level All Flash array (see No compromises for Nimble’s Entry-Level All Flash Array). The new Adaptive Flash arrays provide significant benefits for customers:

  • Lower $/IOPS by 30% to 65%
  • Lower $/GB (useable) by 33% to 43%
  • Increase density in terms of useable GB/U by almost 100% and with deduplication support in an upcoming release, by up to 140%;
  • Increased scalability by 164% from a maximum useable of 440GB to over 1.1PB

An overarching goal of a next generation platform is to drive vastly improved value to the customer. This can be done by upgrading components like processors, and leveraging cost reductions from newer SSDs and HDDs. It’s a no brainer and customers should expect to see improvements in both performance and capacity in terms of $/IOPS and $/GB. But merely replacing components and passing on cost savings is not enough. There is an opportunity with a next generation platform to compound even higher customer value by designing the fundamental architecture to be highly flexible, extensible and tunable from day one. Nimble’s new generation of Adaptive Flash Arrays, the CSx000 (“CS Thousand Series”), is a perfect case in point:




This 4th generation of Adaptive Flash arrays is built on the same foundation and underlying CASL file system (The File System Powering Nimble Flash Arrays) delivered in Nimble’s first-generation Adaptive Flash arrays launched 7 years ago, 2nd generation launched in 2012, 3rd generation launched in 2014, and Nimble’s All Flash array which launched earlier this year.

Performance – A No Brainer for NimbleOS

We’ll start with the no-brainer piece – updating the processor. NimbleOS is the first and only truly CPU-driven architecture for hybrid flash arrays, designed to ensure that generations of Nimble’s Adaptive Flash arrays can simply ride the Intel processor curve and yield significant price/performance gains as processors improve. Unlike other architectures, the performance gains require no additional SSDs or HDDs to fully leverage the CPUs. Utilizing new Intel CPUs in the “Thousand Series” has yielded significant performance gains across the entire portfolio, from the entry-level CS1000 model all the way to the high-end CS7000 model, which pushes the upper performance envelope up to 230K IOPS. The new CS thousand Series shows performance improvements in the range of 33% to as much 192% compared to the previous generation:



 * IOPS based on mixed 70% Read 30% Write


The reduction in the $/IOPS shows significant improvement in performance value to customers:




Packing in Higher Density Storage

And now, for some of the exciting design changes that we’ve made. To achieve our goals of providing dramatically higher storage density, we built a 24-slot chassis that packs 21 HDDs and 6 SSDs (used for cache) into 4U (4U24). A nice outcome for customers is that the 6 SSDs fit into three slots using the same Dual-Flash Carriers (DFCs) that were developed for the All Flash Array (each SSD is individually accessible). Array configurations start with only 3 SSDs and if customers ever need more cache, they can simply pop another 3 SSDs of desired capacity, into the available slots, non-disruptively. The new generation of Expansion Shelves follow the same 4U24 form factor and also come with 3 DFCs (populated with 3 SSDs) with space for another 3-SSDs as an upgrade. A 4U48 All Flash Shelf is also available, allowing customers to leverage up to 48 additional SSDs to massively boost cache capacity if needed.

The CS Thousand Series arrays and shelves support the highest density disks available that are currently at 10TB HDDs, achieving 210TB raw in a single 4U head or expansion shelf, or up to 1.5PB raw in a fully configured system with six expansion shelves.

At such high storage densities, robust data reliability becomes critical. Nimble was the first storage vendor to introduce Triple+ Parity RAID, allowing the CS Thousand Series to immediately inherit the highest-bar of resiliency, even at these high densities. The “+” in Triple+ refers to the addition of intra-drive parity (see The Reliability of Drives) that allows any drive to also recover from partial or sector failure. Along with rock-solid reliability, NimbleOS has been designed to ensure fast rebuild times, in fact the fastest in the industry.

Petabyte Scale

The next decision point was how much scalability we could achieve without the being burdened by the cost of memory (a simple way to think about this is that more memory allows a system to scale to more capacity). The efficient way NimbleOS uses memory allows us to hit our goal of 1PB scale in a single node, across all CS Thousand Series models, AND leave head room for Deduplication and other great features planned for the future. Note, deduplication is already supported on the entire All Flash array line up and the functionality can be leveraged on all new Adaptive Flash arrays. Significant scalability gains of up to 6X usable capacity have also been achieved on the CS Thousand Series, (and this does not take into account any data reduction savings from compression and dedupe):




Data reduction through inline variable-block compression (today) and dedupe (planned for a follow-on release) will yield far higher capacity gains. For example, a deployment that yields a 5X data reduction factor will achieve 500TB of effective capacity on a system in just 4U.

Customers will also see a significant reduction in the $/GB on the new CS Thousand Series platform as compared to the previous generation for maximum capacity configurations:




But wait, we’re not quite done with driving increased value to our customers. The new CS Thousand Series platform inherits all of the benefits of the Nimble Predictive Flash Platform:

Scale-Out – The new platform supports the ability to cluster up to four controller nodes spanning any generation of All Flash or Adaptive Flash. The cluster achieves linear performance and capacity growth and is managed as one.

Scale Up – All models of the new CS array can be non-disruptively upgraded to higher-end models. E.g. a CS1000 can scale all the way up to a CS7000 without any down-time or data migration. (Note that the CS1000H can be scaled-up to a CS3000H).

Variable Block Deduplication – The new platform was designed with the optimal amount of hardware resources, like memory, to ensure the spectrum of models from low-end to high-end will support deduplication that is planned for a follow-on release. Note that the All Flash arrays already support variable block deduplication and the proven code base can be leveraged for support on the Adaptive Flash platforms.

Unified Flash Fabric – Customers can reap the benefit of clustering All Flash arrays with the new CS Thousand Series arrays and manage them as one. Customers who utilize All Flash arrays for primary data can also replicate to cost-optimized Adaptive Flash arrays at a third of the cost (achieving very cost-effective backup, DR and archival for AFAs).

Timeless Storage – The new CS Thousand Series will fall under the umbrella of Timeless Storage which provides for all-inclusive software licensing at no additional cost and guaranteed flat support pricing in years 4 and 5. And, for the first time an Adaptive Flash array customer has the option to receive a free new controller that is guaranteed to be faster after three years.

The Power of InfoSight Predictive Analytics and InfoSight – InfoSight collects millions of sensor data points per array every day in order to predict and prevent problems and keep your systems running optimally. The sensors collected by InfoSight span the entire infrastructure stack – all the way from the storage to the VMs providing dramatic benefits to our customers:

  • Proactively informs and guides without alarm fatigue – Simply put, Nimble arrays just keep working optimally without the customer having to do anything. That’s because InfoSight is always monitoring the array with the goal of predicting and preventing issues before they impact the customer. Nine out of ten issues are discovered first by InfoSight. Surprisingly, 54% of issues that InfoSight addresses are non-storage issues.
  • Predicts future needs and simplifies planning – Predictive analytics and machine learning allow Nimble to accurately predict future needs, hotspots, areas of improvement, and more. In addition, InfoSight has a scenario planner that leverages installed-base learnings to model the requirements of anticipated future workloads.
  • Delivers transformed support experience from level 3 experts – InfoSight automates the tasks handled by traditional level 1 and 2 support staff. This has allowed Nimble to build a support organization made up entirely of level 3 experts who, on average, answer calls in less than a minute. Even the most complex issues are resolved rapidly because InfoSight has already collected all of the information required to solve the problem to root cause. Our storage industry leading Net Promoter Score (NPS) of 85 represents the strong confidence that our customers have about our support.
  • In the future, Predictive Analytics could guide customers to build out the most optimal Unified Flash infrastructure based on the applications being deployed. The recommendations could be determined by correlating application knowledge derived from Nimble’s installed-base with actual measurements of how a particular customer is using those applications. Recommendations could also be made on how a customer could dynamically move applications across the Unified Flash Fabric to achieve a more optimal result.

Nimble’s new CS Thousand Series Adaptive Flash arrays deliver massive improvements in price, performance and scalability to our customers. But their true value is unlocked by the power of the underlying architecture, coupled with InfoSight Predictive Analytics and the Unified Flash Fabric.