This week, we announced the GA of Infinio Accelerator 3.0. More about that in a moment, but first, I’d like to take you back to 2011, the year we started Infinio.
The market, 5 years ago and today
Back then, the vast majority of arrays were almost entirely spinning media. Dell had acquired EqualLogic a few years earlier – which was the beginning of the end of Fibre Channel’s dominance and the subsequent rise of IP based systems. Flash was nascent in the data center, and the hybrid array seemed revolutionary: new companies were emerging, new architectures announced, and entirely new cost models for storage were proposed.
Fast forward five years to now: June 2016. What can we say about the market today?
- The all-flash market is growing at a breakneck pace, outstripping most aggressive projections. Nearly all hybrid companies, many of them young vendors in their own right, have introduced all-flash offerings to stay competitive.
- Hyper-convergence has won the hearts and minds of IT administrators. Talk about a dark horse no one anticipated! What seemed like a branch office or SMB phenomena is now run in some of the world’s largest enterprise datacenters.
- Cloud storage (and public cloud in general) is on an absolute tear. Similar to hyper-convergence, many thought this would be limited to new companies and startups, but the reality is it has become ubiquitous in many industries.
I think it would be difficult to identify an industry in more transition than ours. The fundamental physics of storage completely changed with the introduction of low cost flash in the data center. The fundamental business model completely changed when people started hosting data in the cloud. And the fundamental architecture changed as increasingly more people opted for a movement from centralized arrays and to distributed server-side systems.
But then again, some things never change
I think one of the things we’re most proud of at Infinio is that early on, we chose a few fundamental principles that we believed would never change, no matter how the market shaped up. Even in 2016, I think we can look back and say those have proven true.
First, we believed that increased adoption of solid-state media would make network the new bottleneck.
With our initial product, we introduced the idea of providing storage performance from the server-side because it was faster than going over the network.
Today, that is true in spades. The throughput of NAND flash, DRAM, and future technologies like 3D XPoint completely crush the bandwidth of a network. For example, consider our 3.0 product. We can achieve 1,000,000 IOPS per server, and 20 GB/s of throughput. Forget 10 Gb/s networking—you would need a 160 Gb/s network to sustain that type of load! For so long, the industry was hamstrung by slow spinning media, that when flash became ubiquitous much of the industry momentarily forgot the fundamental rule in data center design: there’s always a bottleneck somewhere, with each innovation, it simply shifts. Today, the network is in the hot seat.
Next, we believed that for an I/O stack to be high performance, it needed to be memory-centric in design.
When flash began to gain traction in the data center, it was easy for many people to forget one simple thing: DRAM is orders of magnitude faster than flash. At Infinio, we knew that the best performance would come from leveraging memory as intelligently as possible, and we did that by creating a content-based architecture with inline de-duplication.
Today, the industry is recognizing that storage and memory have been divided for far too long. The reality is that all of these technologies from DRAM, 3D XPoint, and NAND to spinning media and even tape, are on a single spectrum. In the latest release of our product, we still build around a RAM-based cache, but also enable organizations to tier to flash when necessary to take advantage of its economics while still garnering the benefits of being on the server side.
Finally, we recognized that our customers are busy and run complex environments: Everything about our product has to be brain-dead simple.
I will be the first to admit that adding a server-side storage acceleration solution is not conceptually as simple as saying “storage performance should come from the storage array.” One has to get their head around the idea of separating storage performance from storage capacity. So our goal was to build a product that was practically far simpler than anything else on the market. Our first release was a product that took under 30 minutes to install, requiring no reboots and no downtime, and enabled you to continue using your storage system features and VMware integrations.
Today, all of this remains true of our newest release, except installation is now under 15 minutes, and we’ve earned the VMware Ready certification to further support our customers’ commitment to VMware. It couldn’t be simpler to get going with Infinio, or to live with it after you buy it.
So despite all the change over the last 5 years, from 2011 to 2016—I feel pretty good that we really got a couple of things right, both then and now.
So are you saying Infinio hasn’t had to evolve?
It would be a gross omission to suggest that we haven’t evolved, though, both through our own learning and from the market conditions. I’d say the biggest shift has come from how our customers think about deploying Infinio, and the requirements that puts on us from a product development perspective. I look forward to sharing some of those learnings in a future post.
For now, if you’re interested in seeing the product in action,