Flash Wars Heat Up as EMC and Fusion-io Battle for Top Gun

Quick Take

Flash competitors are aggressively jockeying for position as the market heats up. It’s a tale of two styles. On the one hand, EMC’s entrance into the all-flash array market targets traditional IT segments. It will both pressure competitive offerings and its own high-end block storage business. EMC is positioning to cannibalize its own base before others cut too deep into the EMC muscle; but it must walk a fine line. At the other end of the spectrum, Fusion-io is uniquely positioned to serve the hyperscale market and currently stands alone with a software-led strategy that leverages atomic writes and delivers new value to database workloads.

The pressing questions are as follows:

  • How will EMC’s entrance into the all-flash array market impact existing players?
  • How fast will the all-flash trend eat into high-end block- and even file-based storage solutions?
  • Is Fusion-io’s current software and functional lead large enough to buy it time to enlist ISVs to write to its SDK, thereby locking in an ecosystem advantage. Or will NVM Express find its way through standards committees and actually turn into a product offering prior to Fusion’s cementing its hold on the market?
  • Is the hyperscale market, which is being led by Web giants, a harbinger of future data center design, or a niche confined to the likes of Facebook?

We attempted to address these and other issues by pivoting our analysis off of a recent spate of flash announcements.

The Flash Horserace

Last week several flash players made both strategic and tactical moves, including Violin – new products and a supply chain leverage deal with Toshiba – Virident and Seagate – rumored to be supplying EMC; and EMC’s big flash announcement. As well, Fusion-io made a technology announcement and hammered home its software-led and differentiation strategy.

In his analysis of the EMC all-flash array and new PCIe card announcements, Wikibon CTO David Floyer provided a deep assessment of the nuances of the offerings. While he was very positive overall, particularly about EMC’s all-flash array, he questioned some of EMC’s marketing tactics, specifically its benchmarks against Fusion-io. Floyer’s premise is that while EMC will do very well in traditional IT markets, speeding up existing apps, Fusion-io plays in very different spaces, namely the hyperscale market (think Facebook and other scale out Web companies). Floyer believes that while somewhat narrowly applied today, hyperscale trends will bleed into the traditional enterprise IT space and capture a fair share of the market. In other words he sees Facebook as a harbinger of future data center designs, not an outlier.

Regardless, Seeking Alpha, an open financial blog chose to attribute some specious comments to Floyer and publish some inaccurate statements about the EMC announcement. Floyer addresses those issues in this video interview I did with him today. Namely, Floyer sees meat on the EMC all-flash announcement and factually he clarifies some of the mistakes made in the article.


Flash Market Segments

However, flash transitions are a complicated topic and in the video, Floyer and I spend considerable time slicing and dicing the nuances of this space. In particular, Wikibon sees three segments emerging broadly in the flash market, including:

  1. Flash on Server – 3% of capacity shipped; 20% of spending (by 2015)
  2. Flash-only Arrays – 11% of capacity shipped; 35% of spend
  3. Traditional/Hybrid Arrays – 86% of capacity shipped; 45% of spend

According to Floyer, EMC’s announcement best positions it in the all-flash segment but he questioned its impact on the server-side flash piece of the market. In particular, he was surprised by EMC’s cancellation of its Thunder project and subsequent attack on Fusion-io (the company showed some benchmarks and had some fun adverts de-positioning Fusion).

Hyperscale Markets and Atomic Writes

In particular, Floyer believes Fusion-io is going after the hyperscale market and EMC’s competitive posture toward Fusion was more marketing than substance. Specifically, Floyer believes EMC is really not well positioned for hyperscale markets and its competitive benchmarks were misleading. In his view, Fusion-io’s ability to do atomic writes, combined with its experience in hyperscale, make it the clear leader in that market. Rather EMC is more focused on traditional storage markets, helping clients adopt flash without the need to “rip and replace.”

[Note: Atomic writes are writes directly to flash (like a direct memory access write) without going through a conventional disk protocol like SCSI. According to Floyer, this approach increases performance by 4 orders of magnitude relative to spinning disk].

I asked Floyer “Why can’t EMC do atomic writes?” His response was that EMC and others can, in theory but they are relying on the NVM Express standard to deliver that capability. Floyer thinks the NVM standard is stuck in committees; while Fusion-io is delivering solutions today. Floyer stated that he thinks Fusion has a 2-3 year advantage over NVM Express.

It’s likely that EMC wasn’t using atomic writes on Fusion’s cards and didn’t optimize the Fusion card for performance. Or perhaps EMC ran its XtremeSW software on both the EMC and Fusion-io cards without invoking the Fusion software capabilities—which is like having Usain Bolt run a sprint with his feet tied together. Fusion-io has a rich set of software to optimize database performance and it’s the very high-end database performance where the company shines.

EMC clearly believes it can compete with Fusion using its PCIe card and the XtremeIO all-flash array. This was signaled by the news that it cancelled “Thunder” a clustered scale out version of a PCIe card that was much more like the direction we believe Fusion-io is headed.

EMC most likely feels the market is not large enough to warrant Thunder. Floyer nonetheless believes EMC is giving ground on 20% of the market, the most lucrative high-end, high margin database segment. A key signal here is EMC’s cancellation of Thunder, the shared/protected/clustered PCIe card. It would have been a strong contender in the market but EMC felt the market was too small. Floyer believes that Fusion-io is going hard after the market for hyperscale database and believes it has a big lead. Said another way, Facebook, one of Fusion’s biggest accounts, is an indicator of future data center designs, not an outlier. The hyperscale accounts are bleeding into the traditional enterprise – especially within financial services.

In Summary

My assessment is that EMC’s acquisition of XtremeIO underscores the company’s view that flash is the future and it must control its own destiny. Rather than wait to get hurt by upstart competitors (e.g. a Data Domain redux), EMC aggressively entered the market by making a sound purchase to supercharge its flash business.

At the same time, EMC is facing a tectonic shift in styles of computing. Beyond cloud and big data, two areas where EMC has invested, hyperscale computing approaches are demonstrating the ability to delivery extreme business value. Fusion-io is innovating in this space with a strategy that is remarkable in its innovation and thought leadership. EMC’s strategy, while extremely well-positioned for classic enterprise IT customers is not a direct hit on Fusion’s business. Rather it’s a step in the direction and a move to get in the game.

The key is the speed at which Fusion-io can get ISVs to write to its SDK. I personally believe the big data, NoSQL and open source database crowd – e.g. MySQL, Aerospike, HBase, Cassandra, Mongo, Accumulo, etc. will hop on Fusion’s approach and pressure Microsoft and even Oracle to exploit the Fusion-io method. To the extent that NVM Express is delayed by committee stasis, Fusion-io will thrive. If however NVM can be accelerated and brought to market within the next 12 months, Fusion will be pressured to maintain its lead.

At the end of the day this is all great news for customers because it signals the end of the “horrible storage stack” as my friend Pauline Nist from Intel would say. The more innovation that goes into reducing the mechanical latencies and 40-year hairball of spinning disk, the more value and innovation customers will be able to deliver.




, ,

  • Leo Qiu

    it really undermines the credibility of the article when MYSQL is listed as the first in ” no SQL database crowd”

  • Thanks for the comment Leo. We appreciate the correction – of course we know MySQL is an RDBMS as we deployed our first LAMP stack test instance here at Wikibon in the mid-2000’s – just an honest mistake. I’ve updated the piece…

    It’s important to include MySQL in this discussion because Percona has been particularly aggressive in advancing MySQL in this regard – supporting atomic writes since last summer – http://www.mysqlperformanceblog.com/2012/09/24/percona-server-tree-with-support-of-fusion-io-atomic-writes-and-directfs/

    Personally I feel that this will further pressure Oracle and Microsoft to respond from the database and OS sides and support Atomic writes sooner rather than later. As well I think tech suppliers will respond too. In particular, competitors to Fusion-io won’t (imo) wait for the NVM Express standard and will instead develop their own capability/standard and try to get to market faster than NVM. EMC clearly is a likely candidate to do this – especially given their server-side position with VMware. IBM is also a logical candidate and of course has a single level storage file system in the original AS/400 code base that it can leverage.

    The question is how large is Fusion’s lead and how fast can competitors really develop the capability?

  • Nice article, David (Vellante). My (David Nicholson) nearly objective view is that FIO will have challenges monetizing its thought leadership position in the coming quarters. The market where they have demonstrated success is characterized by a relentless march toward commoditization that chills the spine and the balance sheet. David (Flynn) has hinted that this pressure (possibly from Intel, Pauline?) is what prompted ioScale. The great unknown is how quickly Facebookish architecture will trickle down to “the enterprise”. David (Floyer) is correct that this will inevitably happen “in the future”.

    One thing is for sure. David is right. 🙂

    By the way, we are also looking forward to the end of the “horrible microprocessor stack”. Ask Pauline to please set Atom free. LOL.

  • Pingback: Exadata X5 | Have you hugged your backup today?()

  • Pingback: [Exadata] Exadata X5 | Certificação BD()