In 2012, VMware accomplished a lot by sketching out a vision of the software-defined data center as the direction it's headed. It sees virtualization not as the one-time technology transition early implementers expected, rather as an ongoing process of applying more automation to all forms of data center operations.
Despite the heavy R&D such a vision demands, VMware is on a path to grow revenue at a rate of nearly $1 billion a year. At the end of 2011, it had revenue of $3.77 billion, and it was on course to achieve $4.5-$4.6 billion in 2012. Its rate of growth -- 20% -- was slowing, but still healthy in the third quarter of 2012. If that growth can be sustained, VMware is well down the path toward becoming one of the giants of the technology industry.
But as the scope of its ambitions have increased, so have its risks and commitments. For example, revenues were up in the third quarter, but net income was down -- $157 million versus $178 million the year before. Its annual report warned, "Our current R&D efforts may not produce significant revenues for several years, if at all." Every company has to include this standard boilerplate in recognition of the financial risks it faces. In VMware's case, the risk remains starkly real.
In the coming year, the reach of VMware's goals will require it to juggle some possibly conflicting options and act on key issues to maintain its momentum. Here are five suggestions for how VMware should proceed.
1. Offer Hypervisor Performance Information (Users: Don't Hold Your Breath)
We need more benchmark information on ESX Server hypervisor performance, even though "performance" isn't the only consideration in installing virtualization. The general virtual machine (VM) management environment is certainly as important as baseline performance. Still, efficient execution of tasks is a leading attribute of software, with varying capability often hidden in a set of similar but competing products. If Microsoft, Citrix or Red Hat can claim a performance advantage over ESX Server, it will speed up the painfully slow inroads they've made in VMware's customer base. I suspect the leading hypervisors aren't that far apart when it comes to performance. Some testing indicates that, while other testing shows performance gaps.
More information is needed because there's not a lot of it available yet. When it comes to something as fundamental to virtualized data centers, private cloud operations and public clouds, you would think hypervisors would be so extensively tested that there's little left to learn. In fact not much is publicly known about how well hypervisors perform with different types of applications.
That's because VMware's end-user license agreement prohibits customers publishing their own ESX Server benchmarks. I'm sure VMware would say that's because it can't vouch for their accuracy. It's also possible that VMware wouldn't be able to deny their accuracy. Take your pick. (Another company that incorporates "no benchmarking" in its contracts is Oracle.)
What we do know is that the SPECvirt_sc2010 benchmark produced by the non-profit Standard Performance Evaluation Corp. at one point recently showed open source KVM winning 19 of 27 benchmarks. What might a different benchmark show, maybe the TPC-VMSstandard sponsored by the Transaction Processing Council for database applications? Hopefully that will soon yield publishable results for performance of virtualized databases.
A benchmark published by Virtualization Review in 2009 showed Citrix XenServer, a version of open source Xen, to be "the Porsche of hypervisors," sticking to a fairly fundamental test. Microsoft's Hyper-V was the runner up. VMware dissed the results in a blog on its website, producing this irate and funny responseby then Citrix CTO Simon Crosby.
We need more performance-oriented information. We need to know whether the market leader is also a performance leader or lagger under different workload settings. We need to know whether Hyper-V and XenServer shine in certain settings. And we need to know whether KVM's position inside the Linux kernel, using the kernel's scheduler and memory manager, is actually more efficient or just window dressing.
This is number one on the list for 2013. VMware is not in favor of testers other than itself issuing benchmarks. Don't be surprised to find it on the list again when 2014 rolls around.