BlogProduct » Architecture

Last night Fusion-io made a pretty interesting product announcement, introducing a software toolkit that enables storage networking / sharing of Fusion-io cards in industry-standard servers via Fibre Channel or Infiniband.  You can read the Fusion-io announcement here, as well as a good article from The Register’s Chris Mellor here.  In this post I’ll share a few thoughts on Pure Storage’s take on what this means, and how this impacts the broader flash market.

1. The Tier 0 / Tier 1 segmentation still holds.  Over the past year we’ve been espousing a particular view of the flash market, where existing flash solutions can be largely segmented into two categories, Tier 0 “performance-optimized, server-centric” flash geared at accelerating point databases, and Tier 1 “cost/capacity-optimized” flash, geared at replacing existing Tier 1 disk storage for consolidated workloads.  If you caught EMC’s recent earnings release, you’ll see that they largely agree with us (slide 14).  The Tier 0 play is all about performance and largely serves an enthusiast niche in the market, while the Tier 1 market is all about transforming the existing $15B performance storage market to a modern flash-centric architecture.  Today’s Fusion-io announcement is squarely aimed at Tier 0: enabling Fusion-io to better compete for the performance enthusiast niche.

2. Addressing some of the “inconvenient truths” of host-based flash.  In reality, today’s announcement is really much more of an admission by Fusion-io of some of the challenges they’ve seen in taking host-based flash to the next level of adoption in the market.  There’s no getting around the fact that host-based flash is A) rapidly commoditizing as new vendors enter weekly, B) expensive to deploy due to the inherent inefficiency of per-server deployment and management, C) difficult for customers to adopt given the architectural upheaval it causes, and D) difficult and expensive to protect data stored on it via RAID/mirroring/snapshots/replication, etc.  Yes, it turns out the storage industry got it right a couple of decades ago when we moved to the networked storage model, and the flash vendors are quickly coming to the same realization.

3. Tier 0 flash appliances are still a long way away from being storage arrays – remember it’s all about the software.  If you look at what flash appliances are delivering today (TMS, Violin, Fusion-io ION, EMC Thunder, etc.), they are pretty on-par with storage arrays from the late 90s (basic HA, simple RAID, rigid capacity allocation), just a heck of a lot faster.  But for the past decade the war between EMC, NetApp, et al. has been fought largely on the software battleground, not the hardware battleground – the hardware has become a commodity.  And in particular, for Tier 0 flash appliances to move down-market into the mainstream storage market, they’ve got to deliver software in two key dimensions: high-performance data reduction to make flash affordable (inline deduplication, compression, thin provisioning, etc.) and enterprise-class resiliency and scale to make flash trustworthy and usable (active/active HA, online scale-out and scale-up, dual-parity RAID, snapshots, replication, etc.).  This rich software layer, of course, is what Pure Storage focuses on as its core area of innovation.

4. Build your own…good luck!  Many vendors have tried their hand at enabling “build your own array” software solutions over the years, providing storage management software to run on industry-standard servers for customers to configure with whatever hardware they like (I must confess, I may have even been a product manager for such a solution at one point in my life!).  But such solutions have always failed to make an impact on the market at large.  In the end, it turns out that rigorous system integration, support, HCL, and testing, testing, testing matter!  And in this day and age of shrinking staff and budgets, who has time to run a sideline storage manufacturing business?  It’s just not worth the time and risk.


ION represents a nice step forward for making server-centric Tier 0 flash more usable, but let’s face it – it’s no flash array, and it’s not something that will improve the lives of the mainstream storage user.  Pure Storage’s strategy has remained the same from day 1: focus on the software innovations necessary to dramatically transform affordable Tier 1 storage with flash, and allow everyone to participate in the real flash revolution.

  • techjunkieFLASH

    Interesting take, however I’d argue the following points: “Yes, it turns out the storage industry got it right a couple of decades ago when we moved to the networked storage model, and the flash vendors are quickly coming to the same realization.” – Software ALWAYS follows hardware. And for example. SQL 2012 with ALWAYS ON. You can build a highly available, high performance cluster with all local storage. Having the intelligence of managing the data within the application that uses the data has a significant advantages to using a “mainframe” storage system that just sees blocks. The network storage model came out of necessity at the time by throwing hard drives together. If Local high quality reasonably priced SSS was around at the same time, I’d bet that things would not have evolved in the same way. / “Build your own…good luck! ” – I’d say this is an issue that needs to be handled properly, however the cost savings associated with it, AND the fact that when you look at history, and how ATI/NVIDIA broke the mainframe mold of graphics with a build your own model, the same success can be applied here. And you could technically argue that the “win/tel” combination is similar to what Fusion-io is doing. Its just that its a single company providing the software and “CPU” “win/tel” was successful because it was open, with customer choice of what vendor they wanted with the core elements being the same. iON can do the same.”What about Dell and IBM?” – this is more then likely due to Fusion’s choice in server platform. You can still build your own Dell and IBM iON system. I wouldn’t be concerned.” Tier 0 flash appliances are still a long way away from being storage arrays – remember it’s all about the software. ” I think fusion-io agrees with you. They are making a storage SOFTWARE announcement, and this will enable them to add all the special features like snaps, replication, etc that everyone needs these days.

  • FloppyDriveorDEATH

    I think you may misunderstand what Fusion-io is trying to do. It is not supposed to be a general purpose flash array. Since it isnt, it basically invalidates most of your points. iON is a purpose build solution to address specific software instances where Fusion-io cannot currently provide a solution: Shared Storage. Software is moving away from the requirement of generalized shared storage, to specific localized storage. iON is going to address the legacy software environments such as Oracle RAC which requires Shared storage.

    Fusion-io makes solutions for accelerating the money making portion of a business, not general storage like Pure, EMC, Netapp, etc.. When you buy Fusion, you are buying the DEDICATED performance of a single monster storage array, or a single Flash array, for a single application. And you really only end up spending 10-50k on a any single solution, not 100ks+ for a general purpose array that doesn’t specifically do anything except manage data. Pure’s 100k iOPs is divided up to all your apps and functions.

    Ill be the first to say your approach to a flash based array is the best, but its still general purpose storage. Yes you have dedup, and compression, but technically that just makes your solution a better value than other SANs in certain environments which can take advantage, such as virtual. But a purpose build solution you are not, and that is exactly what Fusion-io is.

    • Matt Kixmoeller

      Hey FloppyDrive, thanks for reading and thanks for your comment, and I think we mostly agree. If you read the Tier 0 / Tier 1 segmentation I’ve had going on this blog for a while, it gets exactly to the essence of what you are arguing: Tier 0 flash players (of which Fusion-io is one) are largely targeted at single app acceleration, while Pure is going for a much more consolidated storage play. We happen to believe the latter is a much bigger market, but it’s clear both are large and viable markets.