Insights, Best Practices and Forward Thinking from the Customer Facing Team, Solution Architects and Leaders on Extreme Performance Applications, Infrastructure, Storage and the Real-World Impact Possible

Why Does An All-Flash Array Architecture Matter More Than Ever?

by Tim Stoakes on January 20, 2015

The beginning of a new year is often an opportunity to reflect and make resolutions. In storage, you may be part of a growing group of customers that has made the resolution to stop investing in disk, and make the move to flash. In this blog series, I will share with you why not all enterprise flash storage is made the same; and that architecture matters.

Violin was one of the first pioneers to see the potential of flash in the enterprise. At that time, even though the cost of flash was in the tens of dollars per gigabyte (which is hard to imagine today), the customer need was clear and some of the largest corporations on the planet were unable to scale their real-time applications on disk, constrained by its inherent latency and IOPS issues.

Today, these customers have deployed Violin to run in environments such as the real-time inventory management for millions of SKUs in retail, consolidating real-time billing records for millions of cell phone users, executing billions of dollars’ worth of real-time daily stock transactions or managing the source code for ten thousand Android developers.

The cost of flash has come down dramatically, and we offer the best performance/price ratio in the industry. With the adoption of inline data-reduction technologies and the continued cost erosion of NAND (including the emergence of 3D NAND), customers are now clearly making the decision to deploy all-flash arrays for tier-1 primary storage.

But what does it really take to deploy flash for primary tier-1 storage?

Let’s first understand what this means: using flash for primary tier-1 storage means the consolidation of a diverse and large set of multiple and mixed workloads on a single platform, with the aim to simplify and reduce the cost of storage management.

There are several implications when doing this:

  • The platform must be able to scale towards petabytes and provide a continuous and non-disruptive expansion path as the data grows in the enterprise
  • The platform must be highly resilient and offer state-of-the-art data protection capabilities, as after all, data loss in this environment is absolutely unacceptable
  • The concentrated set of data makes the platform an ideal place to do inline deduplication and compression, which helps bring the cost of the solution to that of enterprise disk
  • Workloads are getting virtualized or containerized, requiring tight integration with the virtualization or cloud layers
  • Simple management and automation tools are essential

Choosing the right all-flash array that can deliver all of this comes down to picking the right architecture first and foremost.

At Violin we believe the right architecture has the combination of the following 3 product characteristics:

  • The best performance (measured as a combination of minimum of latency with a maximum of IOPS)
  • The lowest possible price/GB
  • The maximum amount of mature software features

flash_image

First, you get the best performance by delivering the lowest latency with the highest IOPS based on the underlying hardware architecture.

We all know that slow I/O holds back total system performance, and clearly you can’t have an all-flash array with “too much performance”. Who would not want to unleash all the power of their servers, reduce application software license costs, and lessen the carbon and physical footprint of their storage to the fullest extent possible?

Second, the lowest possible price/GB needs to be achieved through a combination of driving the lowest possible physical raw cost structure and reducing overprovisioning, and leveraging data-reduction technologies such as inline deduplication and compression.

This is only achieved by making the right system architectural tradeoffs between hardware and software. SSD-based architectures can’t deliver this. You must build your own flash modules from the ground up for the lowest possible raw flash cost and highest possible performance. In software it is about developing an inline de-duplication and compression engine that tightly ties into the hardware flash layer, helping to minimize the cost per gigabyte, but not at the cost of performance and latency.

Finally, software capabilities such as storage data management and data protection need to be all optimized for the underlying hardware architecture and must have been hardened in real life deployments.

While this may sound simple, the reality is that this takes time to get right and properly integrated on an all-flash array: even though some storage vendors have been creatively bundling their heterogeneous storage virtualization products with their all-flash array, this often results in complex and difficult-to-manage products.

At Violin, we believe that we are building the best all-flash array with the right architecture, delivering the right features at the right price. Let’s start the conversation and discuss how we can help you and your company be wildly successful in 2015.

In the next installment of this series, I will talk a bit more about the economics of flash storage versus disk. Stay tuned!