The software-defined data center is a magnificent concept. But any practical implementation would have to include software-defined networking as part of the new, programmable data center operations, and we're not there yet. For VMware to provide it, it will need real networking expertise, but until recently it's been fair to ask: what networking expertise?
Most of the existing networking know-how originated down the road from VMware in San Jose, Calif., where Cisco Systems is located. Cisco provided the first hardware complement to ESX Server's virtual switch in its Unified Computing Architecture and Nexus 5000 switch. It also provided the first (and, for a time, the only) practical implementation of VMware's recommended approach to secure VLAN tunnels, with a VXLAN supporting switch. In this partnership, Cisco has been the senior partner, but both have benefited. Consider the example of Vblocks, the Cisco-VMware-EMC design collaborative that offers preconfigured and virtualized racks equipped with servers, storage and networking.
VMware leads in data center server virtualization and, with EMC as its parent company, it has storage virtualization well in hand. But, because it did not have the expertise to fulfill the networking needed for the software-defined data center, Cisco was able to assume the top position.
That's why VMware's acquisition of Nicira is so important. Nicira represents the next generation of virtualized networking based on the OpenFlow protocol. With Nicira, VMware can map out the network virtualization part and build it into its management console. Nicira's leadership in OpenStack's Quantum networking initiative also allows VMware to tap into the expertise contributed to the entire project. VMware needs this project to succeed to allow switching within the software-defined environment to connect to the rest of the world.
Even Nicira's ability to innovate may be too limited to fulfill all the demands of the flexible networks of the future. It certainly won't be able to fill all the gaps as VMware customers struggle to make the transition themselves. In fact, Cisco, Nicira and OpenFlow startups such as Big Switch have important roles to play, but it's questionable to me how long Cisco will wish to participate in propelling the software-defined data center forward. It's preferred baby steps in that direction, ones that don't threaten its switching hardware product lines.
Beyond that, there's the ongoing issue of virtualized resource management. A software-defined data center will have one management console, not a series of separate ones for the server admin, network manager and storage manager. Without commanding more networking expertise, VMware can't build out the software-defined data center's management console. It won't be enough to have adjustable dials for servers and storage, and oh, by the way, consult your nearest network manager if you don't like the network IT pre-assigned to you.
Changes in the virtual machine will require simultaneous changes in the network, just as it does in storage. When the virtual machine picks up and moves, storage has to move with it, perhaps to be reconfigured at the destination. Can networking do the same? Not today.
VMware must cultivate its own networking expertise and build out virtual network management capabilities in order to keep the concept of the software-defined data center alive. If we're talking about a splendid concept -- and only a concept -- five years from now, something will be deemed to have gone haywire.
3. Decide Whether To Concede The Market's Low End
At VMworld, the most persistent complaint I heard was how much of the IT budget VMware was beginning to command. Customers don't say they're dissatisfied with what they're getting. They say they feel a little bit like addicts, hooked on an expensive drug and wondering if they won't wake up one morning and decide they can't afford it anymore.
There has been tangible value in VMware software. But when it comes to a long-term strategy, I don't think VMware wants to give its customers a compelling reason to consider any of the three lower-priced spreads -- Microsoft, Citrix and Red Hat's KVM -- in the market. Many of VMware's customers already have a toe in the water with one or more of the alternatives. It's not uncommon to hear of VMware data centers with Citrix virtual desktop infrastructure. Or VMware data centers with departments basing local projects on Windows Server's Hyper-V and open source KVM.
VMware somewhat defensively says its total cost of ownership is less than Microsoft's. Last spring it produced a study to prove it. But I think the cost issue turns on what type of infrastructure you want. In some cases, Microsoft's approach will yield the lower cost; in others, VMware's. It seemed false to me to assume that everyone using Microsoft's System Center really wanted to use a VMware-style infrastructure.
That said, VMware conceives of the data center as 100% virtualized and managed from that standpoint. Hence, it comes up with an evolving line of products pushing data center operations in that direction. It wants the hardware to be invisible, or barely visible at the fringes. Microsoft, with its own large virtualization presence, is coming at the problem from the opposite direction: it sees the Windows Server hardware first and layers virtualization on top of it. There's a difference.
But is that difference so great that VMware can consistently charge a high premium? I miss VMware's old days, with Stanford professor and chief scientist Mendel Rosenblum and his wife, CEO Diane Greene calling the shots. At the time, VMware showed craftiness at both charging for its hypervisor and offering a low-end version for free, just as Xen appeared on the scene. Imagine that. VMware dominated the high end of the market, and refused to abandon the possibility of playing in the low-end segment as well.