We’re entering the next phase of the converged infrastructure cycle. The first wave started roughly in 2008 and was driven by a combination of five main demand- and supply-side forces:
- The emergence of consumer clouds and the consumerization of IT,
- IT complexity,
- A rapid uptake of server virtualization,
- An overly labor-intensive IT economy, and,
- Oracle’s 2008 announcement of arguably the first modern computing era converged infrastructure, the HP Oracle Database Machine, a single SKU box which emphasized integration, engineering for performance, and pre-configured systems.
Other major events followed in 2009, including Cisco’s UCS announcement; Oracle’s acquisition of Sun -- which replaced HP as the hardware supplier for Exadata -- the introduction of the industry’s first single-SKU general-purpose cloud infrastructure platform, Vblock from Acadia (renamed VCE, after the Virtual Computing Environment); HP’s announcement of the Converged Infrastructure Architecture; and NetApp’s initial Flexpod announcement.
The race was on to capture share in this huge market.
Check out: Micheal Capellas to Step Down as VCE Chairman.
Disruptive or Evolutionary?
Initially this market segment looked to be disruptive, with Oracle becoming a full stack supplier, and Cisco entering the server market with an integrated network system. The VCE joint venture in particular opened up new markets for Cisco, VMware, and EMC.
However, HP’s counter-punches on Oracle and VCE showed a strong commitment to converged infrastructure from an incumbent server player. Over the past four years, the competitive landscape has evolved, such that today virtually every traditional server and storage player has entered the space, including Dell with its Force10 acquisition, and finally IBM in April 2012 with its Pure line of systems. Even EMC threw its hat in the ring this year with its VSPEX “Proven Infrastructure”, and we’re also seeing some promising upstarts like Nutanix participating.
This market generally is taking shape as an evolution, not a revolution, with more packaging than invention going on. The competitive benchmark has coalesced around real converged infrastructure offerings. The past was characterized by plenty of "solutions," but going from reference documents to deployment required a lot of architecture, testing, and tuning. Today a spectrum of offerings are available, from the rigid (Oracle's "red stack" of Exadata), to single SKU (led by VCE Vblock), to more flexible "reference architecture *plus*" (NetApp/Cisco, FlexPod, and EMC VSPEX). Every major storage and server vendor is now in the mix - see Wikibon’s Primer on Converged Infrastructure. Most have more than one offering, causing some high-level confusion, but the channel typically sorts this out over time based on customer needs.
The competition today is still less around stacks and more "roll your own" vs new stacks; although the exception here seems to be suppliers trying to break Oracle’s stack stranglehold.
IBM’s entrance into the marketplace was a major milestone. In comparing IBM with other major offerings it’s worthwhile to look at Cisco’s UCS. This is a combination of compute + networking that partners with storage – e.g. EMC (VSPEX and VCE), NetApp, and most recently HDS (UCP Select). All of these solutions, HP, Dell, and HP, are multiple products that are tested, bundled and shipped in a rack.
IBM PureSystems, on the other hand, is a single chassis that includes storage, networking, and compute. IBM's solution has more flexibility than most tightly integrated, truly converged architectures, (e.g. variety of hypervisors, Power or x86, multiple vendors of networking, storage virtualization that can front-end other storage), while maintaining a single chassis.
There are two main angles on IBM:
- It is the most broad truly integrated line in the business, competing with virtually everyone, from a reference architecture through Vblock (e.g. with PureFlex) and all the way up to the Oracle Red Stack (e.g. with PureData and PureApplication).
- IBM is late to the market and still hasn’t figured out how to sell Pure. Its channel and sales force are confused, as was, for example, VCE’s in the early days.
To the latter point, IBM made a huge and significant organizational change this fall, placing Andy Monshaw in charge of Pure as a virtual organization. Much like a VCE inside of IBM, Monshaw runs a P&L comprising different parts of the organization. Like Cisco and EMC, IBM is measuring the initiative and going through the process of making sure it can measure success while not double-counting. Nonetheless, Monshaw is a seasoned executive who understands the challenges and will absolutely put Pure on the right track in my view.
Hyper-Convergence and Flash Come to the Market
A new breed of "hyper-convergence" solutions that have compute, network, and storage in a single appliance have appeared. Vendors in this space include SimpliVity, Nutanix and Scale Computing. All of these solutions are designed to be scale-out architectures with the key value being they are software-led. The hyper-convergence players are hopping on the “Software-Defined Data Center” bandwagon and delivering systems that emphasize granularity with independent scaling of compute and storage.
SimpliVity and Nutanix leverage flash as part of the architecture. On December 3, 2012, Nutanix updated its portfolio with more flexible configurations. Nutanix leverages commodity compute and Fusion-io in the architecture. Scale Computing builds a cluster of compute+storage nodes that scale very simply, targeting smaller (0-5 IT people, 50-500 people) organizations.
Flash is relatively new to this space and is starting to have an impact. As mentioned, a couple of the hyper-convergence players are doing flash. Cisco UCS now has options for Fusion-io and EMC VFCache, although volumes are still small. HP has been adding flash into its blade servers that are part of its converged solutions. Note: IBM and HP also partner with Fusion-io.
Use Cases – Cloud, Mission-Critical Apps, and VDI
For most players, converged platforms are essentially Infrastructure-as-a Service. Oracle Exa and IBM’s PureData and PureApplication are the notable exceptions, where the infrastructure is purpose-built and not generalized to support multiple applications across the portfolio. Nonetheless, converged infrastructure generally is an out-of-the-box platform for building cloud infrastructure.
As such the uptake within the cloud service provider (CSP) space is substantial. CSPs are moneymakers and can often cut through the organizational issues many IT shops face (i.e. networking, storage, and server silos). In the CSP world, the decision to go with a single block of infrastructure is more business-case driven than political. We’re also seeing demand from organizations building private clouds internally on top of converged infrastructure.
We’ve also heard from a number of folks in the Wikibon community that VDI is the killer app for convergence. Customers will often use VDI as the first convergence solution and move to other applications (adding it to the initial solution to create mixed workloads). The challenges for VDI have been organizational, with multiple potential owners (desktop, server, storage, network), and cross-training complexities. Similar to general convergence, organizations don’t need specialists to manage the infrastructure, they need an IT generalist, which often causes organizational dislocations.
Finally, there’s a big push for virtualization of mission-critical applications. With the advent of flash entering these systems, we expect to see more mission-critical apps supported on converged infrastructure.
Silobusters or the Next Phase of Lock-in?
The breadth of converged solutions is becoming significant, starting from $25K configs (Scale Computing) to multi-million dollar systems (big Vblock configurations running from $1.6-2.5M per solution). IBM's systems start at $150K-$300K list price, while NetApp FlexPod, EMC VSPEX, and hyper-convergence solutions can all go smaller/cheaper.
Notably, IBM and HP have a multi-family approach - One infrastructure heavy (IBM PureFlex and HP VirtualSystem), the other an application vector (IBM PureApplication and HP AppSystems). The application platforms are purpose-built for specific applications like SAP, Hadoop, Microsoft (and in the case of Exadata, Oracle).
This underscores a dissonance in the market. Specifically, all vendors (with the exception of Oracle) tout these solutions as “silo busters.” That is to say, single blocks of infrastructure that can break the stranglehold and lock-in that purpose-built systems have brought to the market. But shouldn't a standard platform be able to handle mixed workloads? VCE touts Vblock as a single building block that can be configured for the application. That's what Nutanix and others are doing too - not having different families, but understanding workloads, balancing for them, and scaling.
IBM and HP claim to have embedded knowledge into their systems through “Patterns” and “Cloud Maps” respectively. But this creates a danger of evolving highly-stove-piped infrastructure environments into a set of highly converged silos. Better? Yes but will it fundamentally attack the labor content problem – the fact that more than 60% of infrastructure spend goes toward labor? Certainly it will create some improvement, but will it reach the extent desired?
Action Item: Because the convergence trend is ubiquitous, CIOs should be advised to measure business cases carefully. If your vision is to break the stranglehold of lock-in or tear down silos and decrease labor expenses, be careful to squint through vendor marketing and understand how exactly that will happen. As is often the case when children become adults, the exact outcome can be unpredictable.