Building Virtual Empires

We were dining last year with the CEO of a market-leading infrastructure systems vendor, and the subject of a particular startup (that we happened to be invested in) arose. The startup plays a virtualization role in front of this CEOs systems, adding intelligence, flexibility, policy -- but also enabling the substitution of commodity systems.

In short, this CEO was scared. It was a question of owning it, or killing it, he said. “But why?” we asked politely. Could not a mutually beneficial partnership be imagined?

Answer: “No. They suck our brains out.” Presumably this is not good for the systems vendor, but it is a pattern already playing out across the IT strata. Here are some of the implications: If it can be built, it can be virtualized. And if it can be virtualized, it can be commoditized.

The point here is that large shifts are at work in the data center networking field, and these shifts are being driven on all fronts by virtualization.

As we survey the world of virtualization, we find two primary motivations for change. The first is partitioning. This is the realm of innovators such as VMware Inc., probably the single most influential startup in this whole movement, and already a disruptor of historic significance. VMware came to the fore as a champion and agent of server consolidation. Lightly used applications running on low-utilization machines could now be roped together and run on fewer systems at lower operational expense. Other, lighter-weight virtualization schemes at the application layer promise similar benefits in less general cases.Ensim Corp. was an early example of this approach, but a wrong turn into the Web hosting market just in time for its evaporation seems to have set it back a few years. More recently, the little known Twingo

(now known as Cisco Systems Inc. (Nasdaq: CSCO) Secure Desktop post its acquisition) seems to have been on a related technical path.

Incumbent IT vendors with the most to lose are not completely ignorant of the threat posed to them if virtual machines succeed in abstracting their core assets. Microsoft Corp. (Nasdaq: MSFT) moved to acquire virtual machine technology of its own in the form of Connectix, while simultaneously seeking to throw FUD at non-Microsoft virtualization schemes. This is not a new strategy -- virtualization can be a wonderful thing for a vendor as long as it is the one doing the virtualizing. IBM Corp. (NYSE: IBM)demonstrated this long ago with z/VM on its mainframes, a powerful capability that has helped keep the platform thriving for decades.

Beyond Microsoft, other incumbents such as Hewlett-Packard Co. (NYSE: HPQ), IBM, and Intel Corp. (Nasdaq: INTC) are perhaps less at risk today but also see the power at the virtual machine layer, leading to somewhat schizophrenic behavior. On the one hand, they move to enable virtual machine technology (witness Intel’s Vanderpool Technology and Advanced Micro Devices (NYSE: AMD)’s Pacifica). But on the other hand, they quietly work to divide and weaken the virtual machine supply base (perhaps through support of new open-source alternatives such as Xen). From their perspective, virtualization is a good thing in that it expands their opportunity –- but only if the virtualizer is captive or weak. An emerging gorilla in the abstraction layer would pose quite a challenge.

Server consolidation and OS abstraction are not the only reasons to pursue partitioning. Security through virtualization is another important objective, and is emerging as a hotbed of innovation. Early activity appears promising, probably with the best yet to come. As the industry embraces virtual machines, additional opportunities arise to manage them more effectively and use them to solve various thorny systems management problems. We suspect several of these management-oriented schemes will bear fruit.

The second primary motivation for virtualization is aggregation. The theory here is that by abstracting physical elements, the semblance of a coherent whole can be asserted. Single System Image technology certainly falls into this camp, but less ambitious management and provisioning visions also qualify. Such a virtual system has the interesting properties of being more scaleable than any single physical system ever could be, while also being more cost-effective since it can be assembled from commodity elements. The whole argument depends on the competence of the virtualizer, of course. Some (maybe all?) of the intelligence of the predecessor high-end system is still needed, but in the new model it resides not in the physical “atomic” elements (storage, CPUs, bandwidth) but in the virtualizing abstractor. “Brain-sucking,” as our CEO friend from the first paragraph would say.This “aggregation virtualization” theme is in full swing across all sectors of the data center. Network elements, given their strategic positions in the data path, can have a tremendous opportunity in this regard. Examples include Acopia Networks in NAS virtualization and Topspin Communications Inc. in server virtualization. Such in-line virtualization nodes leverage the visibility and scaleability of a wirespeed datapath device, and minimize host-side deployment complexity and server overhead. When done right, they can prove extremely valuable. Cisco has been surprisingly silent on this topic to date, allowing a handful of startups to seize leadership and early market share.

The wheel turns full circle when one considers that the major application architectures of tomorrow are already based on virtual machines. J2EE and .NET are of course the leading examples (with vendor-specific constructs like SAP AG's (NYSE/Frankfurt: SAP) ABAP/Netweaver also on the list). These programming architectures are already built upon a standard abstraction layer. Convergence at the virtual machine presents the opportunity to both partition and aggregate at the same time by offloading the virtual machine processing to a scaleable shared resource that can also offer fine-grained control to its individual tasks.

Azul Systems is an example of a company taking such an approach. The Azul game plan -- delivering optimized infrastructure for an ascendant abstraction layer -- is one that can be executed in adjacent sectors as well. Network Appliance Inc. (Nasdaq: NTAP) did a similar thing in support of Network File System (NFS), a famous successful example. We expect analogous approaches to yield future market leaders in other parts of the “virtualized data center.”

The concept of aggregation leads us into a Wild West of "marketectures" where buzzwords collide: grid, utility, autonomic, on-demand, Linux, blades, multi-core, and multi-thread. The disruptive effects ripple up and down the stack, from silicon to systems, from management to the applications themselves.

Perhaps one of the most important (though least appreciated at the time) watershed events was Intel’s announcement last spring of the cancellation of the Tejas processor project. In truth, Tejas’s demise was really an artifact of a prior failure on another central processing unit (CPU) project known as Prescott. Both were a function of the difficulty of perpetually scaling the clockspeed of the Intel architecture. In the aftermath, Intel, along with Sun Microsystems Inc. (Nasdaq: SUNW) and the rest of the industry, has embarked down the multi-core/multi-threaded path. With clockspeed now taking a backseat, the key initiatives in processor development increasingly revolve around new architectural features that are well suited to the aggregated/virtualized computing schemes of the future. Expect more such 90-degree turns in the other major IT ecosystems.As virtualization takes hold in the major ecosystems, the tip of the industry’s spear aims for a new target. The abstraction layers become the new constants, and take their place as platforms. Traditional elements beneath are commoditized, and later overrun by infrastructure purpose-built to support the new platforms, typically from startup competitors. Such efforts often make good venture investments. If you want additional examples beyond the aforementioned Network Appliance and Azul, consider what happened in the network when IP became the convergence layer -- propelling router companies such as Cisco and Juniper Networks Inc. (Nasdaq: JNPR) to prominence.

Meanwhile, opportunities also emerge atop the new platform to add value, with new vendors usurping the classic roles once filled by predecessors on the old platform. For a history lesson here, take a look at the IBM mainframe. Most anything worth doing was present on that system in the 1970s; the art since that time has been porting these capabilities to new platforms. This cycle has led to the rise and fall of many a sector leader, and quite a few good venture investments. Maybe the virtualized data center is really best described as a “virtual mainframe” after all.

Of course, the greatest upside of all flows from the virtualizers themselves, but these would-be disruptors are fewer in number and also bear the greatest risk. To appreciate the risk, you need to remember that perhaps the greatest engine behind virtualization is commoditization -- for by virtualizing something, you can commoditize it. Harnessing the gravity well of commoditization to create distinctive, high-value businesses is possible but requires great precision in the execution. It’s a little like using a black hole to slingshot yourself across the universe!

Sometimes the virtualizers become lost in the faster current of commoditization, their value propositions not significant enough next to the stark advances of the cheap or free. Other times, the virtualizers simply become commoditized themselves.

The virtualized data center contains all of these opportunities, and all of these risks. The key to winning in this fast-changing environment will be a careful discernment of which virtualizations are long-lived and extensible, and which are needless overheads. With the platforms in view, the rest should be “easy.”— Peter Wagner, Managing Partner, Accel Partners