VxRail An incredible product at an incredible time.

VxRail – An incredible product at an incredible time.

Warning – this is a doozy of a post. For those of you that like moving pictures with sound, tune into the launch here.   For those of you that like juicy details and want to invest time to really understand, read on!

For any offer – any, there are the classic “four P’s” that a team needs to consider. Listening to the market, listening to customers, and adapting is important – we all operate in a very competitive market.   Sometimes we don’t hit the “4 P’s” – and from that you learn.

One needs to hit the “4 Ps” to win in the market, and IMO VxRail nails them.

Pricing – starting at $60K list. This is particularly important in a market which is all about “starting small, and scaling up”.

Product – modern hyper-converged offers need all flash. They need dedupe. They need compression. They need real erasure coding. They need flexible configurations. They need rich local and remote replication. They need cloud object storage integration and local NAS support. They need to make “day 2” operations single-click simple.

Positioning – anyone who says that one converged or hyper-converged offer can cover every use case is (IMO) as high as a kite, or suffering from “single product delusion”. Reality? Some customers can get away with one product for all workloads and scales (for them). Most need more. Most need:

Vblock covers the large enterprise datacenter with the most traditional workloads (and any workload for that matter, but can be over-engineered/over-built/overly-complex for workloads without “classic” infrastructure data service requirements)

VxRack covers the large enterprise datacenter when a customer wants to scale BIG in a simple horizontal scale-out fashion (and are supporting workloads don’t need the most “classic” infrastructure services).   VxRack is hyper-converged and scales easily – but when you plan to scale really big, “don’t worry about the network” is tantamount to planning for failure.

VxRail covers small and medium customer datacenters, departmental enterprises and more. VxRail is built to start small, and scale. Heck with 64 nodes in 16 appliances – it can run 3000+ VMs and could run many enterprises.

VxRail is unbelievable at the enterprise edge – where extending their standards from the datacenter, integrated management, simple workload mobility – are powerful.

Packaging – people who understand technology and innovation as ultimately powering business understand this important.

Only VxRail is a product developed by VMware and EMC operating with a single product team – for this release, for the Q2 update, for what we have in Q3, and for every release afterwards.

Only VxRail is a hyper-converged appliance with a single support model for everything inclusive of the VMware software. VSAN-ready nodes can of course have a single support stack for VSAN and the hardware it runs on, but a hyper-converged appliance is more than hardware + an SDS stack.

Only VxRail has integration with VxRack and Vblock to deliver a full enterprise solution for all workloads, all leveraging VCE Vision, and the ability to replicate and protect from the enterprise edge to the datacenter, and for all workloads. And what we’re delivering in Q1 only will get stronger.

A portfolio matters – as no customer is the same.   Unless of course you are a single-product hammer – in which case everything looks like a nail.

EMC is the CI leader, not by a little, but by a lot. Gartner says it. IDC says it. But with no intentional disrespect – forget those guys for a second.

What’s more important is that the most important people – the customers say it. They say it to the tune of a $3B+ annualized run rate. Our 2015 Q4 revenue growth can swallow up many of the startups annual 2015 all-in revenues. That said – so far, EMC and VCE (the Converged Platform division of EMC) has been all about the “converged infrastructure” system architecture (what Gartner calls “Integrated Infrastructure Systems” in their taxonomy).

What’s going on there?

Well, we have have a ton of happy customers using Vblock broadly – for a ton of data center workloads, including the most classic, the most traditional workloads. When you are operating at material scale, when you have workloads that don’t scale CPU/storage together, sometimes converged infrastructure is the way to go – and Vblock leads the way, and we continue to innovate and improve. EMC and Cisco keep doubling down in that domain. Are we perfect? No. Do we respect the competition, do we listen to the customers, do we always strive, fight, and improve? Yeah. Anyone who thinks they are not in a position to be disrupted is arrogant and delusional.

Interestingly, perhaps our biggest challenge is that Vblock is the “Kleenex” of converged infrastructure – and that people need to know what we are doing in hyper-converged system architectures. Heck people call things “Vblocks” when they mean “converged infrastructure” (lol – sometimes calling Cisco UCS attached to anything a “Vblock”).

We have VxRack customers for data center workloads where a hyper-converged scale-out model is superior to the “step function” model of Converged system architectures.

Now, to be clear, when you are deploying hundreds of nodes at once, I personally would run (not walk) from anyone who didn’t take a full system design that incorporated the ToR switches and Spine/Leaf network architecture (and that enables you to scale to hundreds and thousands of nodes – not deployed in small “clusters”, but as a massive scale-out rack-scale, web-scale, and hyper-scale system.

There are customers who are investing in VxRack that that drive hundreds of nodes (and that’s in a single system, single site) – and growing.   For relative measure – some of these VxRack customers are spending – in a single transaction, single system – the equivalent of ~1/20^th of the annual revenues of the hyper-converged appliance “unicorns”.   That’s a real rack scale system and use case.   Furthermore, the VxRack business is accelerating fast. Are we perfect? No. Do we respect the competition, do we listen to the customers, do we always strive, fight, and improve? Yeah.

What we lacked was a strong offer that covered a critical gap – the ability to start really easily, really small – and then be able to scale up in a very, very simple way.   Not a “white glove” VCE touch, but a “one button touch”.

That’s critical for the Enterprise edge and remote office/branch office. It’s critical for customers who can run their datacenter on hundreds or thousands of VMs. And that “start small, really small” was something we didn’t have in Vblock and VxRack, and our earlier HCIA attempts didn’t get right.

With VxRail we have it, and we think we have nailed it.

Back in 2015, the VMware and EMC senior leadership teams took stock of what we had learnt about this market through earlier efforts:

We learnt that the customers love vSphere, love VMware, and it has become (for most customers – not all) their standard. While there’s great heterogeneity in the Enterprise Datacenter (where Vblock and VxRack play strongly as heterogenous platforms), in many markets (by segment and by use case) – VMware is their standard – period. Customers liked the promise of taking the platform they have selected and making it easier.   That was good!

We learnt that trying to design software and hardware separately didn’t work when you are talking about a system value proposition. It’s not that the hardware isn’t a commodity (it is), but rather you ended up with configurations that were too fixed, not sufficiently integrated, and an ability to innovate in both domains that was constrained (software that lagged, hardware that lagged, simply due to a complex co-engineering model).

We learnt that customers wanted far, far more flexibility in configurations than we originally thought. We thought that simplicity (and at the ultimate end of simplicity, you have a single part/single type of node) was paramount. The customers corrected us. They wanted simplicity sure, but they also wanted more. They wanted flexibility in configurations. They wanted flexibility in maintenance and support (versus a fixed 3 year support model). They wanted to bring their existing ELAs to the party.

We learnt that there was a critical set of features and capabilities that if you don’t have – you’re just not competitive.

We learnt that this is a market that doesn’t have a a lot of pricing elasticity. In fact, one of the biggest factors we discovered was entry price point. The “lowest entry price point” is a huge deal.

Solving these problems required a different approach than prior efforts.   Materially so. Step 1: we brought together the VMware and EMC teams and made them one.

If you look at the launch of VxRail, you’ll see many, many moments where it’s apparent that when it comes to VxRail, VMware and EMC are not a “loosely coupled Federation”, but instead are singular.    Literally, the VxRail team is single organization, completely integrated – who have now delivered more than 200,000 engineering hours together – and are only getting started.

Their charter, their mission is simple: build, deliver, and rapidly iterate to create the market’s leading hyper-converged infrastructure appliance – bar none. That HCIA offer must be the gold standard for customers who have made vSphere their standard – and to make it the simplest, easiest system to deploy, scale and maintain.

VxRail is that offer – and is a quantum leap forward, the first of several quantum leaps for 2016.

VxRail integrates with VxRack and Vblock for a full converged/hyper-converged portfolio that is without peer.

With all this tight partnership, what about an open ecosystem? While VxRail is something unique to VMware and EMC and our only Hyper-Converged Infrastructure appliance offer, it’s important to recognize that VMware maintains and open ecosystem for Hyper-Converged software elements – vSphere, VSAN, VSAN-ready nodes.

Likewise, sometimes customers want/need heterogeneity (common in large datacenters) and EMC offers VxRack and Vblock which can run any hypervisor, bare metal stacks, and container/cluster manager models.

That is important – customers want choice. And of course, the server ecosystem are absolutely welcome to build on top of VSAN ready nodes and add their own value.

But – when you want a turn-key appliance solution, VxRail is the answer from VMware and EMC.

There will certainly be naysayers. There will certainly be some that point to past efforts (you live, you learn, you adapt, and you iterate forward). I’m SURE that I’ll have a pile of competition pile on – and I respect them all (and respect the large and the small equally) a great deal.

Ultimately – the only proof is in what the customers, and the market says, but I’ll reinforce my personal point of view:

Customers want partners with a portfolio that covers their full spectrum of use cases, inclusive of blocks, racks, and appliances.

Customers want partners with them for the long haul.

Customers want partners who can provide solutions that run above the converged/hyper-converged infrastructure stack – full turnkey EuC/DaaS, IaaS, PaaS, and Data fabrics.

Customers want partners who understand that technology and innovation is critical, but also that support, maintenance and thinking about the big picture is a pre-requisite. Uniquely, and I want to be clear here – while there can be ecosystem support models for hardware and software in things like VSAN-ready nodes – when it comes to a full appliance experience – the simplest, easiest model – with full single support is VxRail – there is no other appliance with “one stop” support for the full software stack if you want to have vSphere as your core virtualization layer. All others are “stretching” if they make that claim. Full stop.

Let me fill you in a little more re technical details of VxRail. Read on!

The first thing to understand is we classify VxRail as an appliance – with a picture, above!

Note what’s not in the picture – rack & networking. When we say “appliance” that means totally turnkey, with a “bring whatever rack and networking infrastructure you have” type of offer.

It also means modular scaling in building blocks. Like what you get from one?   Need more?   Just add another appliance.

This is the more common hyper-converged offer that people think of today, and I think (?) that for example everything that Gartner has in the Hyper-Converged Infrastructure System (HCIS) segment falls into this bucket. “Start small, scale and update easily” is the mantra.

Sidebar: Hyper-Converged Appliances are distinct from what we would classify as a “Hyper-Converged Rack Scale System”.   I mentioned this earlier, and want to elaborate.

This “Rack-Scale” part of the system taxonomy is not as well defined. While still hyper-converged in the sense that they use software-defined storage stacks, and scale in a small “step function”, they are different than appliances.   If “start small, scale and update easily” is the mantra from appliances, then for this class of solutions “scale BIG, scale, operate, and update easily at BIG scales” is the mantra at Rack-Scale.

Rack-Scale systems must incorporate the network fabric as a core part of the system design and management stack (which means they start bigger than appliances – physically and economically) – why? Simple, they are designed to scale at “rack scale” or greater. Rack-Scale systems also commonly use a much wider set of industry standard server hardware, often in “disaggregated” or “composable” infrastructure designs.

Note that in Rack-Scale systems, it’s not enough to just “bundle” the network, like is common with hyper-converged appliances. The network needs to be part of the system design – and is influenced by workloads, the nature of the SDS stack, and the nature of the abstraction/pooling degree of horizontal scaling.   The network needs to fall into the coverage of single support, lifecycle, warranty – and of course management integration – i.e. everything that goes with the networking domain being integral.   None (!) of the Hyper-Converged Appliance offers does that (and that includes VxRail). VxRack is that type of Hyper-Converged Rack-Scale system taxonomy – but that’s a story for another day – today is about VxRail.

Let’s get back to VxRail and Hyper-Converged Appliances.

When you look at VxRail, it uses the very popular, very dense 2U4N (2U form factor, 4 module) form factor – and ultimately will be able to start with 3 nodes, and scale in 1 node increments. Note that while we’ve started with the 2U4N form factor we will continue to expand VxRail to support a broad range of form factors over time.   More on the hardware variations later – but unlike prior efforts, there is a ton of variability in configs.

That’s a lot of power in a small package (as I keep telling people – should never, ever be underestimated 🙂

Think 112 CPU cores. Think 2TB of DRAM. Think 76 TB of NAND flash. Think 8x 10GbE ports – for more bandwidth than can be imagined for many customers.

VxRail can scale easily – and frankly support a ton of workloads.

With 16 appliances (which translates to 16×4 nodes = 64 nodes) a scaled up VxRail cluster could support 3000+ VMs, with millions of IOPs and 1.2PB of all-flash capacity, raw (before we apply dedupe and compression). That’s a lot of oomph.

BTW – if you KNOW in advance that you’re going to scale that big, you’re better off to start thinking about rack-level scaling, pre-plan the spine-leaf network and consider a VxRack to start – it means a bigger initial step, but when you start getting to node 100 or 1000, you’ll be thankful.   But I want to be clear – if you’re not sure how big you will scale – no worries, start small with VxRail and grow, and tackle the ToR and Spine/Leaf network domain when you need to. At the top ends of the scaling curve, you may regret having not pre-thought about the fabric that links the racks (because you bet it will matter at that point) – but you won’t be in trouble.

Now, when I said that we learnt that customers wanted more configurability, I meant it.

Ultimately this configurability manifests in the following hybrid models:

… and the following all-flash models:

Not all the models GA at the same time (some almost immediately here in Q1, some arrive a little later in Q2), and there will be a series of software updates (to all configurations – including VSPEX Blue customers) that add features and functions – always non-disruptively.

Now – I personally think that the all-flash models will be preferable to hybrids.

I think we’ve reached the tipping point that for any transactional workload (and VMs are transactional workloads), all-flash configurations are the sweet spot – not only because the dedupe, compression and other data reduction features are designed for flash – but simply because we have hit the inversion point, where there’s a small number of arguments for hybrids, the arguments all-flash (power, cooling, IOps, Latency, bandwidth, predictability, lower failure rates) – all lean to all-flash.

This is true of external SAN models (XtremIO, all-flash VMAX), but it’s also true of transactional SDS-based configurations like VSAN too.

A huge part of any hyper-converged infrastructure appliance is the SDS layer – because in material ways, it defines the system behaviors and envelopes.

VxRail is a natural extension for customers how have said “look, vSphere is our standard”. When you are focused on vSphere and vSphere uniquely, VSAN is the strongest SDS – bar none.

Virtual SAN 6.2 is face-meltingly awesome, and to understand what it can do – you should read this, and this.

Ok… You back?

Great – now you understand the power of all the new all-flash leverage, the data reduction capabilities, the inherent advantages of a kernel-integrated SDS target code layer and more.

This isn’t just my PoV.   Storagereview has done a very interesting set of benchmarks and comparisons of kernel-mode SDS stacks:

VSAN here – “VSAN performance was stellar and that remained true as the data footprint grew” – and that was before 6.2, and not all-flash!)

…And ScaleIO/VxRack Node here – “we’ve never seen any HA shared storage perform as fast as ScaleIO has in our lab. ScaleIO has shattered records in every situation so far, pushing our testing infrastructure to its limits”

…and they also compared them with userspace/guest space SDS stacks used in other HCIA stack (I won’t link those in the spirit of “always stay positive”).

I wrote earlier that I didn’t see the delta on userspace vs. kernel mode SDS stacks, and I was right in some domain, but wrong in others. I’ll say it – I was more wrong than right.   The big impact of core parts of the SDS stack being in the kernel is on CPU utilization on the host, in some cases latency of operations, but most of all how the overall system performs as the CPU utilization on the hyper-converged host starts to climb materially.   That last one is important. Who wants something that starts great, but then starts to act wonky as you start to use it in the manner you intend?

But – what if in addition for the storage for your fast VMs, what about if you wanted some NON-TRANSACTIONAL storage?

What if you needed some NAS for CIFS shares, or NFS (general purpose, or for an NFS-based datastore?).

Well – historically, one downside of hyper-converged appliances is that for capacity-oriented, non-transactional scenarios – frankly, they haven’t been too hot. It’s hard to get the right density, relative to compute, and certainly compared with non-hyper-converged design points (like ultra-dense storage arrays).

So, we thought creatively – and day one, VxRail can leverage almost any S3-compliant cloud object store for cold VMs, Archives, as well as any general purpose NFS, CIFS or iSCSI storage needs – and every single appliance comes with a 10TB license to do exactly that. You heard right. 10TB license – and you just need to provide a cloud object store.

Of course – this can be any S3-compliant object store – a pretty big set of choices.

If you’re worried about security, it’s all encrypted, and compressed. Not good enough, or want a really economical option?   Then use the best, most cost-effective S3-compliant (plus a whole heck more) object store of your very own – Elastic Cloud Storage.

No other hyper-converged appliance does that.

What about the question of local and remote replication?

Surely VxRail, like most Hyper-Converged Infrastructure appliance offers has the same somewhat limited set of data services in this regard – long RPO async (minutes) generally as the industry watermark? Nope. Yes, if that’s what you need – sure.   But, how about incredible enteprise-class replication technology, inclusive of consistency groups, VM-level replication, sync, low-RPO async, continuous replication – and of course with great simple ease of use, and the industries best compression and deduplication across the WAN? RecoverPoint for VMs is included (licensed for 15 VMs per appliance – and you can easily add more), and completely integrated in the UI VMware administrators call home – the vSphere Web Client.   No other hyper-converged appliance does that.

What about rich data protection – that integrates with what the customers already have, versus forcing new backup models on the customer?

VxRail has embedded data protection, but can also replicate to other VxRail appliances.

But what’s even better is that for customers with larger datacenter requirements that have selected the leader in the purpose-built data protection platform, Data Domain – VxRail has turn-key integration with Data Domain that enables simple point and click backup and recovery that leverages DDBoost to offload the network, dedupe all the content, and make it simple to have “edge <-> core” data protection that is simple and just works, and works with what you already have.   No other hyper-converged appliance does that.

What about simple stretched clusters over short distances? Check.

Net?

VxRail:

Extends your standard. If you’re using VMware, and you’re a small/medium business, a small/medium enterprise (remember 3000+ VMs can run many a datacenter!) or a large enterprise with edge/ROBO locations – look no farther.

Is ridiculously simple.

Starts really small ($60K list).

Scales really big (64 nodes & thousands of VMs!).

Scales and does all “day 2” operations with the click of a button.

Performs better than any other Hyper-Converged Infrastructure Appliance on the market particularly as CPU utilization ramps up and you actually use what you bought vs. working about workload contention with the SDS stack.

Rocks and rolls in all-flash configurations, and man, in 2016 – why not go all-flash?

Has rich data reduction services (dedupe, compression)

Has efficient erasure coding that saves you money, and doesn’t have all sorts of caveats.

Has unique abilities to leverage cloud object storage, add local NAS

Can extend your standards from edge to your core:

For support/consumption/health integration with VCE Vision (just like VxRack and Vblock) – an area where there is simple integration day 1, but count on us doubling down and iterating quickly.

For disaster recovery (with enterprise-class replication capabilities like consistency groups, sync/async/continuous RPOs, VM-level protection, single click integration with your STANARD management UI (vSphere web client).

For backup – leveraging your existing standards, and not forcing new backup investments. If you have Data Domain, it just snaps right in.

While today is all about VxRail, in the end (certainly with EMC and VMware) it’s about a portfolio because customers are all unique. Ultimately, there is NO other Converged/Hyper-Converged Infrastructure provider who could actually deliver on the picture below.

Now, as always. Are we perfect (and – to be clear – that extends to the whole picture above)? No. Do we respect the competition, do we listen to the customers, do we always strive, fight, and improve? Yeah.

I know the EMC team (inclusive of VCE as the converged platform division) and we have a passion for the customer. The whole team and I will keep fighting, keep pushing to improve on all fronts with our customers front and center in our mind.

In fact, the voice of the customer and the partner is always the most powerful voice – and the right way to end this post.

Consider these examples – and each of them are people who have broad industry experience with not only our technologies, but with the industry broadly (including our competition).

The voice of a partner: “The addition of VxRail into the VCE product family will allow us to now provide our customers with a comprehensive portfolio of converged systems for almost every use case– vBlock and VxBlock for Tier 1 workloads, VxRack for Tier 2 dev/ops environments, and now VxRail as the ideal remote/branch office offering.” – World Wide Technologies

The voice of a customer: “VxRail offers us real choice; we can scale from our trackside environment to a complete data center solution all from one vendor.

Our trackside infrastructure travels around 75,000 miles per season, supporting Renault Sport Formula One^TM Team with continuous IT services at every race worldwide. The small footprint of the VxRail is desirable because our infrastructure must work in extreme temperatures, limited space and other challenging conditions. We are also pleased with the solution’s ability to support our critical workloads and the masses of data generated each week without any compromises on availability, power, and speed.

I can already appreciate the cost and time-saving benefits we could gain by deploying a condensed, self-contained, standard 2U hyper-converged rack using best of breed storage and compute technology. For trackside, the savings in freight charges alone would be approximately $200,000 per season, and the simplified set-up of VxRail means we could be up and running much quicker at every race, offering a significantly improved user experience.

Hyper-converged infrastructure offering continuous availability, speed of access and time-saving features, and all in a self-contained 2U package which just works – VxRail’s pretty smart.” – Antony Smith, Renault Sport Formula One, IS Infrastructure Manager

….that all says it more than I could.   The feedback we’ve gotten over the last few months during the beta program has been priceless, and is that the root of my passion for VxRail.

Thank you and congratulations to the VxRail team for this great day – and I know that while you’ll briefly celebrate, I also know your passion to be awesome, and you’ll be right onto the next software update right away (which all VxRail customers will get gratis, and will be able to apply non-disruptively).