Rolling Review Kickoff: Server Virtualization

Let's play a word association game. One condition: Pretend it's early 2007. Ready?

"Virtualization."

Did you say "VMware"? Last year, probably 10 out of 10 IT workers would have responded with that one name. If you were talking enterprise server virtualization in 2007, you were talking ESX.

What a difference a year makes. The question of whether virtual machines are here to stay has been answered; now the questions on IT managers' tongues are, "What should we virtualize? And which virtualization product makes sense for us?" Questions of market presence, performance, compatibility, and cost are usually tossed into the mix as well.

While VMware is still the market leader for server virtualization, the company's data center dominance is no longer a given. Citrix has aggressively repackaged its sales pitch and business model, metamorphosing from a terminal services and application delivery provider into a full-service virtualization vendor in an incredibly short time. Microsoft introduced Hyper-V as an integral part of the Windows Server 2008 product line, shaking up pricing models and generating deep thought as purchasing managers assess the ROI on large-scale VM proposals.

Smaller players such as VirtualIron and Parallels are further stirring the pot, offering full-fledged hypervisor solutions that compete with the big three virtualization vendors in niche markets, offering unique products addressing the needs of small businesses and non-Windows shops.

Virtualization has gone mainstream: Business users pull prepackaged corporate VMs over the WAN; college kids rely on VMware Fusion or Parallels to run Windows on their Macbooks. IT shops recognize the potential operational benefits in large-scale virtualization; CFOs are hoping to get a double win from capital savings and reduced server-farm power bills. But what realities lurk behind the hype?

Do the full feature set and mature ecosystem of VMware justify the price tag? Does Hyper-V address the physical-to-virtual consolidation goals of most system admins? Does the open source flag-waving of Citrix, VirtualIron, and other players create a robust hosting environment via "natural selection" and the support of thousands of Xen community members and developers?

There's only one way to answer these questions -- a head-to-head Rolling Review on the same hardware, using the same testing methods, and relying on the same business and technical requirements.

Physical To Virtual
We'll be starting from the ground up in our test scenarios. We will build four VM hosts, two identical higher-end servers reflecting new purchases, and two less-powerful servers representing repurposed equipment freed up as part of our VM server consolidation exercise. All hosts will run a bare-metal hypervisor, with guest VMs and data sets stored on an iSCSI SAN.

The reviews will evaluate each vendor's ease of setup, configuration, and data and network connectivity. Each vendor's physical-to-virtual conversion tools will be used to migrate real-world servers running Windows Server 2000, Windows Server 2003, Windows Server 2008, Windows XP, and Debian Linux into the virtual world. We will clone instances of the resulting VMs to generate up to 10 guest servers per host at maximum load.

File servers, LDAP directory services, a centralized accounting application, an Apache server, a legacy facilities management system, and a Postgres database will be run on respective guest VMs. Our goal is to reflect a typical physical-to-virtual exercise, where large numbers of low-utilization servers are chosen as candidates for virtualization. This reflects our experience in the real world as well as consistent feedback we get from readers and analysts.

We will balance guest VMs across host servers according to vendor recommendations, implementing live-migration capabilities to move VMs from host to host as load and business rule sets dictate. We'll rate the comparative utility of VMotion vs. XenMotioning vs. Hyper-V's pause-and-move solution.

We also will be implementing ill-advised migration rules and triggers to see how our test systems respond under duress with poor guidelines in place.

Simulated user and application loads will be automatically generated against each virtualization platform. We will assess acceptable loads for differing hardware resources within a single vendor platform and ultimately compare like tests across different hosting environments.

Finally, we'll evaluate bundled management tools for ease of use and functionality, assuming that users are experienced IT administrators new to the world of virtualization.

(click image for larger view)