It’s an exciting day for Pure Storage, and one that is full of “3s”: we’re introducing our new 3rd-Generation FlashArray, the FA-400 Series, we’re introducing along with that a 3rd-generation of our software, the Purity Operating Environment 3.0, or “Purity 3.0” for short, and we’re celebrating 3 years of production maturity, with now 100s of FlashArray units shipped, the longest-running FlashArray customers have now had solutions in production for 3 years. Our CEO, Scott Dietzen, penned a blog today to talk about Pure’s market progress, as well as introduce you to our new mission/vision – Flash for All. I’d encourage you to read that first if you haven’t yet. Here in a two-part post, I’m going to dive deep into the product news. Buckle up – as this is our most substantial product release since our first, these are going to be long, detailed posts. This first post covers the new hardware, and a second post covers the Purity 3.0 Software.

New FlashArray FA-400 Series Hardware

Almost exactly a year ago we introduced our FA-300 Series, and we’re back a year later introducing the FA-400 Series, which boasts double the performance and double the capacity of the previous generation. The FA-400 features Intel’s Sandy Bridge processor architecture, and represents a substantial improvement in about every technical specification:

If you’d like to read all about the details, you can download the datasheet, or visit the tech specs webpage.

But ultimately, hardware’s just hardware. It gets faster every year, and so does the FlashArray. More importantly, this hardware refresh highlights a few critical architectural and performance advantages of Pure Storage, which I’ll dive into below.

The Power of Commodity Architecture: Moore’s Law + Rapid Iteration

Broadly speaking, there are three flavors of all-flash products on the market today:

  1. Legacy disk arrays retrofit with flash
  2. All-flash appliances, where raw NAND is coupled with core IP delivered in hardware, burned into custom ASICs and FPGAs
  3. All-flash arrays, leveraging standard x86 processors and off-the-shelf SSDs, with their core IP in software

The first category isn’t really worth discussing, so we’ll skip it. Most of the leading legacy vendors plus all of the newer flash-centric startups have picked a position in either category #2 or #3 – and this is where much of the competition in the industry is today. For our part, Pure Storage’s strong belief is in model #3. We think computing’s history is on our side, and by betting on an open, software-driven approach to innovation, we are able to ride two powerful forces in the industry:

  • Moore’s Law – or more broadly the Intel x86 architecture. Intel delivers to us a doubling of processor power every 12-18 months, the key ingredient to delivering a faster and bigger FlashArray every year.
  • The commodity SSD market. The SSD market is evolving even faster than processors are, and the result is that every 6 months or so a new batch of SSDs come to market that are bigger, faster, more reliable, and less expensive.

The Pure Storage FlashArray was architected to be able to take advantage of both processor and SSD advances quickly and incrementally, enabling us to rapidly upgrade the performance and capacity of the FlashArray. Take a look at our 4-year history, we believe this trendline will continue year-after-year:

Those who innovate in hardware face challenges on two fronts. First, hardware innovation is simply slower. Early approaches leveraged FPGAs to be able to iterate a bit faster, but as flash lithography compresses the advanced ECC technologies required to manage flash now require dedicated ASICs – a slow development path. Secondly, most of the vendors who innovate in hardware simply don’t also build the expertise to deliver rich system-level software as well, and so invariably the hardware-centric devices feel like exotic flash chassis where somewhere along the way the software was left behind.

Pure’s strategy is simple: we leave the hardware innovation to others, take advantage of the rich processor and SSD hardware marketplace, and we focus all our engineering effort and IP around being the software leaders in flash. Over time we believe this market will follow the well-worn path of many markets before us: the hardware commoditizes, and it’s all about the software.

Incremental, Online, “Evergreen” Upgradability vs. the Flash Toaster

The second key advantage that today’s release showcases is how the FlashArray’s architecture was designed to be able to deliver incremental upgradability, or a new “evergreen” storage model, as the Pure Storage founders like to call it, vs. the closed, non-upgradable “flash toaster” appliances of our competitors. Upgradability has long been a challenge of the storage industry – the common model is to buy a disk array, use it for 3-4 years, then throw it away. This was already wasteful in the disk storage world, but in the flash world it is even more problematic:

  • Flash evolves quickly: customers want to be able to take advantage of flash advances that come every 6-9 months, as opposed to getting stuck with the flash of 3+ years ago.
  • Start small and grow: as flash is new to most customers, many want to make a small initial deployment (maybe a single database, or a VDI pilot), but would like to be able to incrementally grow and expand their flash investment over time without ripping and replacing.
  • Downtime isn’t an option: flash tends to be deployed in more important, Tier 1-type applications, and for those applications, downtime for array maintenance simply isn’t an option.

So, Pure Storage delivered an architecture for the FlashArray that has separate, stateless active/active controllers, and independent storage shelves, all of which can be deployed and upgraded non-disruptively: i.e. without user downtime or performance impact. As such, the FlashArray can grow from the smallest 5-10TB usable configuration, up to 100+TBs usable, all without wasting hardware investment or taking user downtime.

Pure Storage makes it easy to start small, dip your toe in the flash water (we promise, it’s warm!), and expand as you growth more comfortable with flash. Want to see how upgrading works? Check out the second post to see some live demo videos.

No More 4K IOPS Vanity Benchmarks!

You might notice that in this release we’ve chosen to start quoting FlashArray performance in 8K IOPS instead of 4K IOPS. The previous-generation FlashArray FA-300 was spec’d at 200,000 4K IOPS, and the new FA-400 carries a spec of 400,000 8K IOPS. We did this to start being more aggressive in our assertion that the flash industry is waaayyyy to over-focused on 4K IOPS, which very rarely exist in the real world. In fact, the average IO size delivered from our global pool of customer FlashArrays (averaged across all the call-home data in our database) is roughly 38K, you can see the live number on this page. 38K!!! People are often surprised by this number, as they’ve been lead by other vendors to believe that 4K IOs are commonplace, and often people confuse their application’s and OS’s block geometry vs. actual IO size. Even if a database is set to 4K IO alignment, multiple layers in the IO stack are happy to ask for more than one 4K block at a time, leading to many larger IOs.

So…why is this important? It highlights an important architectural difference between Pure Storage and much of our competition. Most competitive flash products (Violin and XtremIO, to name two), are architected around a fixed 4K internal block size, which means that if they receive an 8K IO they break it into two 4K IOs, if they receive a 32K IO they break it into eight 4K IOs, etc, which leads to a linear drop-off of IOPS by block size. Pure Storage, on the other hand, implements a variable internal geometry, meaning that we analyze IOs for data reduction down to 512-bytes in size, but we store a single IO as a single IO. That doesn’t mean we’ll have the exact same performance for a 4K IO as we’ll have for a 32K IO, but we’ll be substantially sub-linear in our performance fall-off. Check out the difference:

The result is that for IOs in the “zone” of most real world customer workloads, Pure Storage will substantially out-perform the 4K-optimized competition. Try it out and you’ll see!

Better Efficiency than Brand X: It Takes 4 X-Bricks to Equal One FlashArray

Finally, let’s compare the new FlashArray to the long-announced-yet-still-not-GA Brand X. Based on publicly-discussed information from EMC on Brand X, each x-Brick is planned to be 7TBs usable after RAID/metadata overhead, from a base of 10TBs of raw flash per brick. Let’s assume that since both arrays are deduping/thin provisioning arrays, most customers will want to run the arrays with about 20% empty space, to allow room for safe levels of thin provisioning over-allocation and potential changes in the dedupability of data. An important difference to understand is that Brand X does deduplication only, while Pure Storage does both deduplication and compression inline. Analyzing the call-home data from all the FlashArrays at customer sites, the FlashArray delivers on average 5.6-to-1 data reduction, of which the compression savings portion is 2.1-to-1. This means that if we assume that both arrays will get similar deduplication results (a kind assumption, since the FlashArray can detect duplicates down to 512-bytes, 8x smaller than Brand X’s fixed 4K chunks), then whatever those dedupe results, the FlashArray will additionally reduce data by >2-to-1 with compression, resulting in less than HALF the ending space usage on Pure Storage. Additionally, Brand X’s hardware require dedicated standby power supplies (SPSs) and dedicated redundant InfiniBand switches, all of which consume space and cost. Here’s what the result looks like visually:

The result: a single 23TB Pure Storage FlashArray can store more data than 4 XtremIO x-bricks, and can do so in less than half the physical space (10U vs. 23U).

Ready for a Test Spin?

Well, if you’ve made it this far, you are clearly interested. Our FA-400 Series FlashArrays are available and shipping today, we’d love to give you the chance to experience on in your environment. Click here to arrange a private demo or evaluation!