Kernel-based Virtual Machine (KVM) is an open source hypervisor, which is code that manages operating environment interaction with underlying processors. It’s used to create virtual machines (logical machine images) on x86 servers. Dozens upon dozens of these logical images can run on x86 servers, enabling them to do more work by keeping the processor busier than running just single images, leading to significantly improved server utilization rates. Instead of running at 5 to 10 percent of capacity, x86 servers with virtual machines can be configured to run at up to 80 percent of capacity in some configurations.
IT market research firm Gartner predicts that, by the end of 2012, there will be 58 million x86 virtual machines. Gartner further predicts that:
• EMC’s VMware, the market’s leading x86 virtualization environment, should have approximately 65 percent market share.
• Microsoft’s Hyper-V is expected to own 27 percent of the x86 virtualization market.
• Citrix Xen is projected to have a 6 percent market share.
Meanwhile, KVM’s 2012 market share is projected to be only 2 to 4 percent (Gartner suggests 2 percent while market research firm IDC suggests 4 percent). Note that KVM’s market share is comparably small because KVM came to market almost a decade after VMware.
So here’s the big question: given KVM’s miniscule market share, how can it be argued that KVM will ultimately erode VMware’s solid market position?
In 2009, Microsoft’s Hyper-V virtualization product offering was estimated to have an 11 percent market share. But, in the ensuing years, Microsoft greatly expanded its Hyper-V product offering, adding new server-attached storage, live migration, and site-to-site disaster recovery. Improved functionality, plus low cost (Hyper-V is available for free on Windows Server 2008) has led to its solid growth. As a result, Hyper-V grew 62 percent last year according to IDC.
The same arguments—low cost and greatly improved functionality—are now working in open source KVM’s favor. KVM is packaged along with Linux kernel implementations. Because it’s open source, it’s available for free. Further, in May 2011, BMC Software, Eucalyptus Systems, HP, IBM, Intel, Red Hat, and SUSE announced the formation of a consortium known as the Open Virtualization Alliance (OVA), an organization that fosters the growth of the KVM virtualization ecosystem. The more than 250 OVA members have helped greatly expand the KVM ecosystem by adding hundreds of new infrastructure and management products to the KVM product stable. The open source community is also focused on expanding the KVM ecosystem with hundreds of “oVirt” developers, working to expand the KVM stack.
VMware vs. KVM Functionality and Costs
IT buyers look first at functionality, then cost. In terms of basic virtualization features, VMware and KVM are remarkably similar. In an InfoWorld hypervisor/stack comparison, VMware and KVM closely matched up in the following areas: bare metal provisioning, headroom on virtual CPUs per host, maximum addressable Random Access Memory (RAM), page sharing and memory over-commitment, VLAN support, live migration, load balancing, availability, shared multi-pathing, shared resource pools, and availability of an Application Program Interface (API). There are indeed differences in product functionality. VMware offers live storage migration, while KVM doesn’t. However, live storage migration code is available from the open source community, so IT administrators can download live storage migration code and integrate it with KVM. Overall, these products accomplish much the same tasks.
As for cost, there are major differences between VMware and KVM. VMware generates revenue by license sales of its VMware infrastructure and management products, and from support revenue. KVM has no licensing fees; revenue comes solely from Service and Support (SnS) fees.
VMware and Red Hat SnS fees are almost identical; license fees are what drive up the cost of VMware.
The VMware Pricing Faux Pas
On July 12, 2011, VMware shifted from pricing its virtualization based on physical resources to pricing based on virtual resource usage (this is known as Virtual Random Access Memory [vRAM] pricing). The VMware community erupted. Users pounded VMware with complaints, many illustrating how VMware had suddenly greatly increased their computing costs. (A litany of these complaints can be found at http://communities.vmware.com/thread/320877?start=0&tstart=0.) In short order, VMware was forced to revise its pricing, doubling the amount of vRAM its customers could use and, by doing so, better aligning prices with customer budgets. At VMware’s recent VMworld 2012, VMware again changed its VMware pricing scheme; this time completely eliminating vRAM pricing, instead linking the price of VMware to the number of CPUs on a physical system.
It can be argued that VMware’s initial vRAM pricing action represented a classic case of vendor lock-in, a scenario where a vendor has established a firm foothold in various enterprises and then ratchets up the price of its goods and services because the cost to switch away may be too prohibitive (considering retraining, increased risk, lost time, etc.). It can further be argued that VMware’s vRAM pricing showed its customers how vulnerable they were to its pricing dictums. As a result of this vulnerability awareness, several VMware customers formulated back-up plans that may involve migrating to KVM over time.
What Users Are Saying
Price is important to IT buyers, especially in emerging growth countries and industries where budgets are tight. Consider a Linux administrator (identity withheld) from the University of Nevada, Las Vegas, who was interviewed at Red Hat’s recent Red Hat Summit: “I just couldn’t find the funding for VMware,” he claimed. “I had a department that needed more processing power, so I deployed KVM on their Linux servers to increase capacity. It works well and cost me nothing.”
Price is also a factor at Dutch Cloud, a major KVM and VMware user. In a recent teleconference, Martijn van Zoeren, CEO of Dutch Cloud, a leading Internet Service Provider (ISP) based in The Netherlands, indicated that Dutch Cloud was seeking an alternative to VMware due to its cost. Dutch Cloud implemented the KVM hypervisor for virtualization, and augmented it with IBM SmartCloud Provisioning. The KVM hypervisor, as expected, significantly lowered virtualization license costs. But what surprised Dutch Cloud was the big bump in performance they got from using KVM. IBM has integrated KVM with its x86 server solution (System x) and is now reporting that KVM on System x has demonstrated 18 percent better virtual machine consolidation performance vs. the competition (meaning VMware).
At IBM’s Research Compute Cloud (RC2), users that can run their applications on Linux might opt for KVM based on a lower chargeback fee. IBM’s RC2 is IBM’s largest internal cloud deployment; IBM engineers use it to conduct cloud computing research. This cloud uses more than 200 IBM iDataPlex servers (large x86 rack systems) with about 2,000 concurrent VMs operating at any given time. This environment uses both KVM and VMware for virtualization. According to systems engineers who support RC2, KVM is especially good for deploying virtual machines in Linux environments. It appears to perform better than VMware, though hard data on VMware performance is tough to obtain because VMware’s End User License Agreement (EULA) prohibits customers from publishing such data without scrutiny and prior approval from VMware.
KVM is positioned for strong growth over the next few years—driven by price, competitive functionality, and the growth of a strong ecosystem. VMware’s price model has changed to adjust to competitive pressure from both KVM and Microsoft’s Hyper-V, both of which are free add-ons to Linux and Windows, respectively. But VMware’s license costs continue to position VMware as the high-cost virtualization solution.