[UDPATE: 1/28/2015: 8:13pm PT - Some people have taken some erroneous impressions from this post, so I want to make a couple things clear:
- Poop flinging is bad whoever does it. I'm not claiming EMC is innocent (I thought this was clear, but some clearly missed it - see comments - so I'm calling it out right up front)
- Poop-flinging is different than debating architecture, and when a customer presses you on what you think about someone else, I think it's fair to share your opinion on architectural strengths and weaknesses. Every solution has pros and cons, and ultimately a customer needs to approach their decisions with a high degree of vendor skepticism (only reducing that investigation when they have a real proven track record of being a trusted advisor). Debating architecture vs. Poop-flinging - it's a case of degree. You know poop flinging when you see it and smell it :-) Also, I **AM** saying that people should be VERY skeptical about information provided by vendor A regarding vendor B, as it's usually tainted, and wrong.
- Some think I'm saying all this from a "lets all be nice" Canadian perspective. Not at all! In fact, I smile overtly when I'm in a competitive situation and someone flings poop - because that's the first step of me utterly, and totally destroying their campaign, and their credibility, and ultimately winning - and getting the chance to be a trusted advisor for the customer. Reality is that the market is fiercely competitive. Poop-flinging is fundamentally a BAD STRATEGY - whomever does it.
I want to open the way I will close: this title above applies to us at EMC (me too!) - frankly anyone.
I also don't want to "bury the lead" (as I tend to do in longwided posts), so here's the lead:
- AFAs are the way to go for most transactional workloads - and this is more true every day.
- XtremIO 3.x is doing great!
- XtremIO 4.0 is coming, it's making something that is already incredible even better, and it's 100% non-disruptive.
- "Xpect" :-) to see lots of poop flinging - stand clear, and try to avoid it at all cost.
Read on for more details (and to find out what the heck triggered this post :-)
It's absolutely amazing how fast the industry moves. As much as going on in "platform 3" (open source, SDS object stacks, PaaS models, Tachyon/Spark and other in-memory models) - the same rate of change is happening in "platform 2" land (the vast world of applications which depend on infrastructure resiliency).
The fact that the "platform 2" world is so much bigger (in terms of workloads, and in terms of revenue/profit pools - not necessarily in terms of impact on market/humanity) is why there is so much interest in the new world of All-Flash Arrays (AFAs).
AFAs are a radical "step function" in the persistence layer for transactional workloads - and the price/performance/time curves on NAND are moving SO fast - that in just a few years, we've landed at a point where AFAs are THE way to service transactional workloads.*
*That simplistic sentence has as usual a bit of of an "it depends" list for those with the intellectual honesty and stamina for "full completeness":
- AFAs are not right (at least not yet) for workloads that require "classic enterprise array" data services like huge at scale replication (think tens of thousands of objects replicated, extreme replication topologies, active remote copies, thousands to tens of thousands of consistency groups). There are no AFAs that support this (yet). However, many customers who consolidated workloads on arrays that do those things, only use it for a subset of workloads. Keep those workloads there - and move the rest onto AFAs.
- AFAs are not right (at least not yet) for workloads that require mainframe support.
- AFAs are not right (at least not yet) for cases where a customer needs "one thing to do it all moderately well" - including backup-to-disk, object/cold archive, rich enterprise NAS. I want to be clear - AFAs are not reserved for "Tier 0, or Tier 1" anymore. Between NAND economics and data services - they can be used for any transactional workload. BUT sometimes customers literally need things beyond transactional storage and need it all in one thing. This is the land of VNX, of NetApp FAS, of Nimble - etc.
- AFAs are not right (at least not yet) for the "oceans of object storage" or for traditional (vs. new transactional and in-memory) batch HDFS use cases. It will be interesting to see new hyper-dense, very low write use cases of NAND appear. They aren't here yet - but might be soon :-)
I want to "make it real" in terms of how much the AFA category is eating the transactional storage world:
XtremIO will be the fastest product we've ever done that hits a billion dollars a year. Tune into our Q4/2014 results call (January 28th) and find out when.
If XtremIO was a startup (vs. part of EMC) that grew that fast and continues to accelerate, the market would be speculating that this startup was going to massively disrupt EMC, and speculatively would assign a $4B valuation (after all, NetApp does $6B in revenues and has a $12B valuation - and people are assigning strangely constructed valuations to AFA startups). I would humbly suggest that the stock market does not know how to value EMC :-)
For perspective - that's WAY faster than Isilon. That's WAY faster than Data Domain. That's WAY faster than VMware. That's not to take anything away from those (and other) extreme growth examples - they are all amazing. Rather - it puts an exclamation point on XtremIO's success. You can see those visually on a graph here.
And - perhaps most importantly, customer satisfaction on XtremIO is through the roof.
Clearly we're doing something right. Customers speak. The market speaks. Everything else is just talk.
Here is another external voice (wikibon).
Take a look at what they observe about the cost (all in) of NAND relative to magnetic media (based on very real IDC data). They also (ouch for all vendors) do a great job of getting down to the cost for pre-data-reduction GB by looking at the macro shipped capacity vs. published revenues (note all vendors do data reduction differently, and as you can imagine, that's a big factor).
Take a look at their observations about the tipping point (hint: now) for transactional workloads.
Considering how much debate exists over what is an AFA and what isn't, it was also of interesting note for me was how they classified the generations of AFAs (gen1 = use of SSDs in existing platforms; gen2 = hybrids that optimized for flash.; gen3 = purpose-designed AFAs that were "dual controller"; gen4 = scale-out AFAs). IMO it's a step too far to assign "good/bad" attributes - for any given customer the answer can vary - but there is a clear progression.
I'll point out that the AFA competitor I personally think the most about is Solidfire because architecturally - I think they have the most legs (and are a gen4 design using the wikibon grouping). I don't spend a LOT of time thinking about it (as we're clearly in and winning in the AFA space, and are a gen4 architecture).
I more time thinking about the Hyperconverged market, our friends at Nutanix - and our entry into that market. Like the AFA market - for the last few years, the market was uncontested for the most part. That will be changing very, very shortly. Competition is good for the market, and good for the customer.
Back to the work of AFAs and XtremIO.. For a mental picture of the installed base, we have about 2/3rds of our XtremIO population (which is growing by leaps and bounds) on the 3.x codebase.
Aside for competitors going negative (and some isolated customers with whom we did not communicate sufficiently clearly - to whom we have apologized and made amends) - response from customers has been overwhelmingly positive to the 3.x bits, particularly how we were transparent and stood by our customers and partners in the 2.x to 3.x upgrade.
If you missed all the (false) drama a couple months ago in September when we did the 3.x release, read more here. It's interesting and instructive that it was one of the most commented posts I've ever had. Odd, no? Take a look at the comments to see the voice of competitors relative to the more important, but quieter, voice of the customers/partners.
There's more good news coming on the back of 3.x. XtremIO 4.x is coming soon!
- 4.x will be completely non-disruptive for customers on the 3.x codebase (as I noted in my post on the 3.x upgrade).
- It will introduce the #1 desired feature - non-disruptive X-Brick addition into a cluster.
- XtremIO has always been a scale-out architecture (which is better than being limited to a single HA node), but not a dynamic scale-out architecture. This was always in the design (even at the time of acquisition) - will be great to have it as a fully exposed and completed capability.
- ONE reason why this is good for customers is the ability to reach higher performance (1M+ in a cluster) and capacity (effectively 1PB+ in a cluster) to be sure - but the bigger reason is more subtle. AFAs are SIMPLER than traditional designs - you put a workload on, and it's happy. 4.x will make it simple to add more capacity and performance. The linearity hallmark of XtremIO (under load, with data services) will have another bonus - which is when you need more, you simply add another X-Brick. Done. This is the part of the "simplicity" value proposition of Isilon for scale-out NAS - but only really works with a massively distributed persistence model.
- Native RecoverPoint support. This will add the industry's most popular replication product, with rich consitency groups, fan in/out, broad RPOs, rich data replication, and WAN optimization to XtremIO. There will be no other AFA on the market with that sort of replication. BTW - people that know Recoverpoint well - understand that this is integrated with the XtremIO snapshot engine, so there isn't a "write splitter/journal". This means that there are some things different from Recovepoint CDP (yes, this is a hint that RecoverPoint itself will soon have two modes of operating - continous/jounralled and snap/ship). We have customers already using RecoverPoint and XtremIO together - but this is a big improvement. Previously, to add RecoverPoint to XtremIO required the use of VPLEX which was good in one sense (adds active/active behaviors), but also added cost, physical space, some additional operational complexity and an additional design consideration (IOPS maximums of the VPLEX engines)
4.X will introduce these all these capabilities on existing XtremIO array hardware. No need to upgrade controllers or purchase a new array.
. Plus lots lots more (come to PEX for the details - I'm not kidding, it's a BIG release!)
Nothing is perfect - and XtremIO isn't. Additional things we're working on are rich QoS models (often desired in the service provider segment) - expect to hear more on that soon. That said - with 4.x what we think (and the market is showing) is the best and fastest growing AFA on the market for enterprises of all sizes will get even better.
Now on to the second part of the title of this post. People going negative.
As you might expect - some of those with things to lose if XtremIO succeeds will to do whatever it takes - including going negative. I want to call out "some" - because I have many respected colleagues at all competitors who fight honorably, and I will compete with all day long and have cocktails and dinner at night. I ALSO want to call out "some", because (and this kills me), sometimes I see my EMC brothers and sisters doing things I don't support or agree with. Poop flinging isn't a "company culture thing", but rather a "people culture thing".
I was saddened to see an email today from a former EMCer who left EMC for an all-flash array competitor. I won't say who it is, or where they work, or speculate publicly on why he left, but I have my theories.
Was disappointed to see an email from him to a happy XtremIO customer who chose EMC over their product suggesting that the upcoming 4.x upgrade was destructive and required new hardware. In fact, he even used my transparency in my public blog post on the disruptive 2.x - 3.x upgrade as an effort to suggest that this would be the case for 4.x Not true.
I'm sad for, and disappointed in, my former colleague, but I can imagine he's under a lot of pressure - and we're not making competing with us easy. D - you know who you are. If you want to talk - drop me a line.
My suggestion - for partners and for customers:
The AFA market is hot, and fast and furious. In any hot market like this - you can see lots of poop flinging.
Heck, look at the Rocket/Docker poop flinging recently. I think the CoreOS team has a technical point with a great deal of merit, but the way they did it was tacky and classless, and the Docker response was class. It makes me root for Docker, even though I technically agree with the CoreOS team (but see Docker working to correct the ecosystem point)
Here's a list of things to consider making part of how YOU operate:
- Distrust anyone (ANYONE) who starts talking negative about the other guy, particularly in describing what they do/don't do.
- FACT: It's hard enough to stay current on oneself, and focused on how to build solutions ourselves. The likelyhood that the person flinging poop is full of poop is high.
- Ask your partners/vendors to describe what they think they do WELL, not what the other person does poorly.
- SHUN (absolutely SHUN) vendor created comparison tables (everyone has them, including EMC - I'm trying to make that a poorly-regarded practice). you'll recognize them by one vendor being all "check" and the other being all "exes". It's intellectual laziness. Sometimes the customers demand them. Please don't be an enabler :-)
Note - this isn't new, this has been part of the Presales Manifesto for years (here - principle #5: "be a positive force"), and people know that any "how to compete" training I've ever delivered FOR YEARS has been to delight when a competitor goes negative (and guide firmly to not do it ourselves!). More often that info is wrong (information disparity) - and the trust with the customer, the partner and the ecosystem goes to pot.
Who are YOU hearing go negative most? (customer/partner voice strongly desired). If you are an XtremIO customer - please share your experiences, and thoughts on 4.x? What do you want us to do better?