Flame Smack! Our story about Nexenta at the VMworld 2011 Hands-on Lab (HoL) created a frisson, more than that in fact, and an impassioned EMC exec who was quoted in the story sent in this rebuttal comment. It's forthright stuff. Here it is:
Disclosure – EMCer here
Chris, I know you need to create high drama to drive views to your content. I understand.
Heck, that's what drives readership of the Sun in the UK, the "phone tap" scandal, and the Tea Party movement in the US. It's not good, scientific, fact-based, or useful, but hey – I understand.
I do think that, it's **laughable** to start with a thesis – driven by one of the storage vendors. I don't even know what to call that kind of journalism. The Nexenta CEO (who as you note formed the basis of this article) is well – incorrect.
All of the vendors underneath VMware at the HoL have one simple mission: be invisible. That means work flawlessly, and make sure that we do our small part to give the 14,000 labs and 140,000 VMs created/destroyed during the week a good experience.
Any of your readers who have actually pulled something like this off know that in practice, pulling together, and pulling off an event like this is fraught with difficulties.
When we sat down with the VMware HoL team and started planning for 2011 – we decided to come loaded for bear – and that was the config that we brought.
1) The total load (at peak) could have been serviced by a single VN5700. Look at the processor utilization on the VNXes as shown here (YouTube video) (this was during a peak load period in the lab).
I'll also point out that it seems to me that one of the vendors here has come out, providing details, screenshots, results, YouTube videos, in response to pithy quotes. It's also notable that while I have the data on load, latency across the environment – my blog posts and content didn't say "Hey, vendor _____ hiccupped here, here and here". I guess while some folks want to view little guys as the David in a David vs Goliath battle... when you are the leader, it doesn't behoove us to go negative, but I guess it does for others.
I suggest getting pricing from one vendor on another (like you used in the basis of this article) is ... well kind of silly :-)
2) The primary driver to have TWO arrays there was simple – it gave us emergency options, if one of our arrays was damaged in shipping, if components failed at the show, heck, if we needed to carry the full load – FOR WHATEVER REASON. It also enabled VMware to mitigate risk during the labs themselves. IMO – that sort of "think things though" and "customer first" (the customer here being the VMware HoL team) approach is something people should look for from their technology partners, and perhaps not so much CEO quotes like you got there from other folks.
That's also the root of my comment about "street prices for a given solution are about the same". Compare any "two head system on COTS hardware" and 160 NL-SAS spindles and a small amount of flash to a similar config (which also uses x86 hardware and SAS-connected enclosures) – in this case it would be a VNX5700, 160 spindles, and a couple of SSDs as FAST Cache. While our hardware is based on commodity x86 as well (on all EMC platforms), customers often dig the fact that our hardware enables them to get a lot.
Examples of what we get still out of our hardware include density (60 drives in 4U today, or 25 2.5-inch spindles in 2U) and bandwidth (4 lane, multi port SAS) in enclosures as an example. Extreme HA (high availability) in components in the storage processors is another example.
But, the fact that we use all "off the shelf" component parts mean that all three of us (EMC, NetApp, Nexenta – heck everyone) are driven by the same market forces (both on the cost side and customer budget side). Every customer has choice – and I stand by my comment, and would encourage any customer to get quotes and come to their own conclusions.
Chris – conversely I suggest getting pricing from one vendor on another (like you used in the basis of this article) is ... well kind of silly :-)
Get a quote on a VNX5700 system with 160 NL-SAS drives and a couple flash drives configured as a read/write cache costs and compare it to a similar NetApp or Nexenta system (or any competitive platforms for that point). EMC's continued market success is a reflection of customer value. It applies at the microcosm (every customer choice) and at the macro (overall market trend) levels. We are less than perfect (aren't we all), but strive for every customer, every partner to be the choice on our merits, and at the macro level, that is reflected in results.
As the leader in a market, yeah, it's maybe fun for people to hammer us - and that's OK. That's the price of being the leader. Anyone who views us as dinosaur, well - let them think that :-)
Not a sexy story perhaps for you – but some facts
Fact: EMC has fully embraced a "flash 1st" approach to storage models – and actually have been shipping longer, and in far, far greater volume than any of the primary storage vendors. We use flash as a cache, as a tier in primary storage devices and also have clearly indicated what we're doing around server-side flash. The only other vendors in that are shipped and is supporting flash in a similar league are not storage vendors - but server vendors, or PCIe-based flash folks.
The point that the commoditisation of flash is actually the real "storage" story here, is based on the fact that all the storage configs would have been very different without flash. Almost all vendors have moved to commodity hardware. In my opinion, the "real story" (not just the storage story) is the ability to actually provide a lab cloud that supported that dynamic a load, live – at a major event. Oh, and it used public vCloud provider partners. Interestingly the vCloud partners used in the VMware HoL also use EMC storage by the way.
But, back to flash: we all would have required insane spindle counts to support the workloads without it (again, that was my out of context quote about why that's the main takeaway – thank you very much!).
@Jim 59 (referring to a comment to the original story which said: "Why does nobody mention dedupe in VMware stories ? The storage is basically holding 1000 copies of Windows 2008, right ?")
By the way, that's why dedupe (while an important technology) isn't the main point here. Yes, there were loads of VMs with similar content – but the main design challenge here was not providing the necessary capacity (GB), but being able to handle the very bursty IO loads (IOps). Efficiency is about BOTH $/W/space per GB and per IOps (and per cost of unplanned change).
Fact: You won't get an accurate view from any of the storage vendors (including me) on the VMware HoL – since we each see our piece of the puzzle. The only people with an accurate view are the VMware HoL team – since they were "on top", and in the end, it was their application we were there to support. What I will say is that EMC did our part (and more) to make the VMworld HoL a success. I'm also glad that VMware has a partner vendor community where there are many who help them achieve their ongoing market success.
Fact: What is inaccurate is the whole story. And, perhaps, just perhaps (after all, who am I to tell you how to do your job), starting with a premise based on a Nexenta CEO world view is well ... wrong. His facts are incorrect, and ultimately the VMware HoL staff know.
Fact: At EMC, our technology, our customers, and our results speak for themselves.
After all – it may not be a sizzle story, but it's damn sexy to many to say "EMC technology is market-leading, is gaining market share based on customer choices, and, as a company, is massively investing in today and tomorrow".
Fact: When it comes to storage, those storage technologies of today and tomorrow are: server/storage hybrids, all-SSD designs, mixed storage/server use of SSD, scale-out Block, NAS and Object storage models. More importantly than any of those component technologies, [they are] overwhelming solutions integration with the most important use cases (VMware most of all – but also Microsoft, SAP, Oracle and others).
At EMC, many of those technologies are shipping in volumes that shock, and many we are demonstrating [have] advanced research and development projects in many places. We have leadership not only in share, but in these core technologies and integrations with the applications that ultimately drive infrastructure choices.
I personally welcome competition – from any and all angles. It helps us stay sharp, helps the market and helps customers. We continuously change and adapt to extend our leadership position.
Fact: It may not be a sizzle story, but to the people behind the HoL (and the users) – and to customers every day around the world – it's damn sexy to say "it worked", and I will say that the EMC parts of the infrastructure were designed well and performed their duty in an exemplary manner – and that's why customers overwhelmingly choose EMC.
Take that, El Reg! ®