Today, EMC hosted an event to update the market on its 2016 flash strategy. We praise EMC’s drive and market leadership around flash – more so than any other legacy storage vendor, they’ve taken an attitude of going “all-in” on flash, and are making it very clear to customers that flash is where it’s at in 2016 for any production workload – a point-of-view that we can’t help but agree with, since we’ve been touting it since the turn of the decade!
Before we get into the details of EMC’s announcements, I wanted to start by acknowledging a few trends which I think will make flash even more interesting in 2016. We call these the “Big 3 Flash Disrupters:”
- Affordable flash is here. It’s been here for a long time in fact, but there’s really no denying it today. With AFA pricing down to or even under $1/usable GB, there’s no point in discussing price anymore. This stuff is available for mainstream applications.
- Cloud is driving increasing flash adoption. With all the major cloud vendors building their core architectures on an all-flash foundation, if you are building a cloud (public, private, hybrid, or otherwise) – you need to be doing it on flash. Flash is key to driving higher consolidation and mixed workload performance that makes the cloud model viable. And as the core “cloud” application model moves from VMs to containers, this space is going to get even more interesting.
- Data is changing. The early 2010s saw the rise of “Big Data,” where organizations found value in centralizing their data resources (aka “data lake”), and driving a wide range of analytic insights. But unfortunately, in this era, “Big Data” almost universally meant “slow data”— as batch-style analytics were common. With the rise of in-memory databases, copious amounts of flash and new, more transactional analytic platforms like Spark and Kafka, a new wave of data analytics value is going to rise.
So amidst all this change and given their announcement today – is EMC well-prepared to ride the Flash Disruptors of 2016?
Enter the Lamborghini: DSSD
After years of speculation, we’ve been eagerly awaiting the DSSD announcement. We have tons of respect for Bill, Andy and the whole DSSD team – from the early days of DSSD we often competed with them for key hires in Silicon Valley, and it was clear they were taking on an ambitious project.
Let’s start with what we think DSSD got right: a bet on embracing raw flash, a focus on next-gen data sets, a simplification of the layers, and a common object store. While increasingly larger SSDs are affording the storage market the ability to continue to plug flash into existing storage architectures, we can’t help but get excited when we see products built from the ground-up to really take advantage of flash, and that’s clearly DSSD – built for going after Disruptor #3.
Where – in our view – does DSSD fall short? The software and the scale, making it a really fast RAIDed JBOF, connected with proprietary wires, HBAs and software interfaces. At Pure we believe that hardware innovation only takes you so far — the real magic happens in software. In particular, software tightly-coupled with hardware and engineered together. Not joining this compelling hardware with a complete storage software stack was a missed opportunity: no data services, no copy services, and no data reduction.
That said, it’s fast, and it will unlock some new types of applications and analytics that aren’t possible on other flash products. But at what cost? There’s a long, troubled road of failed extreme-performance “Tier 0” solid state appliances and cards that have struggled to gain significant market traction over the years: Texas Memory DRAM appliances in the 90s/2000s, Fusion-IO and PCIe card adoption en-masse, Kaminario’s original DRAM appliances, Skyera’s direct-connect appliance, and EMC’s very own Thunder, killed just 4 years ago by EMC for being “too niche.”
The challenge with each of these approaches were three fold: hardware that was too finicky and exotic to reach real enterprise stability, super-high cost and limited scale, and a requirement for too many changes to the application model via custom interfaces. In short – a product for the storage 0.1 percent, kind of like a Lamborghini. Just looking at DSSD’s launch references makes this clear: super-computer centers and the DoD.
Return of the Prodigal Son: VMAX3
We are scratching our heads over this one. After a two-year run of pushing XtremIO anywhere and everywhere (particularly into the VMAX installed base), EMC seems to have reversed directions and is demoting XtremIO in favor of VMAX again. In fact, XtremIO barely got a passing mention in today’s launch event. We have already seen this behavior change in a number of customer accounts over the past month – and this makes it official.
I’m not sure any outsider will ever know the reasons as to why, but we can only speculate on some combination of:
- XtremIO potentially not demonstrating the level of availability that VMAX customers demand (remember: always test HA and NDU performance under load in PoCs!)
- A lack of enterprise replication and availability features in the XtremIO platform
- A growing frustration from the EMC field that selling a more efficient XtremIO into a VMAX account depressed revenue and possibly profit for both the sales team and the mothership
But this switcheroo leaves customers with a confusing and frustrating choice:
- Pick VMAX3 when the highest availability is important, and price and efficiency are not
- Pick XtremIO when price and efficiency are important, but I can tolerate some downtime/disaster recovery compromises
And herein lies the problem: Resiliency is always important, but if I am serious about building a cloud environment for my applications, so is cost – and clouds just aren’t built on VMAX economics (or complexity). Today’s announcement was also very unclear in what actual modifications were made. How is this not simply yesterday’s VMAX filled with SSDs, and packaged in pre-configured configurations bundled with software and slightly more favorable business terms?
Finally, as we wrote on Friday’s preview blog, EMC|Dell have more than a few all-flash platforms at this point – nine completely Independent all-flash hardware and all-flash software OS stacks, to be exact. No consolidation came with this announcement. The reckoning is coming. Significant consolidation is needed. It isn’t a reasonable or viable business strategy to maintain nine platforms – I’d suggest you can probably cover the entire market in three or four, max. If you are buying into any part of EMC or Dell’s flash portfolio – you should be asking your account teams the tough questions – are you committing to a multi-year relationship with a product that is on the wrong side of the accountant’s line?
Teeing-up an Exciting 2016
EMC announcements aside – we couldn’t be more excited for the the year ahead in flash. If you want to hear more about the Three Big Flash Disruptors – you can bet we’ll have a lot to say about them, and a few surprises up our sleeves at our Pure//Accelerate conference in just two weeks. Come join us!