Virtualization Shoot-out: Citrix, Microsoft, Red Hat, and VMware
In data centers large and small, the move to server virtualization seems as unstoppable as the waves crashing on the beach in Hawaii's Waimea Bay. But for almost as long as the virtualization tide has been rising, there was only one vendor that could offer the features, interoperability, and stability necessary to bring virtual servers out of the skunkworks and into daily production. That is no longer the case.
From the beginning, VMware has been the king of x86 server virtualization, hands down. VMware's feature set, reputation, and pricing all reflect that fact. But where there used to be little competition, you'll now find a select group of challengers that have brought a wealth of enterprise features to their virtualization solutions and begun to give VMware a run for its money.
[ Also on InfoWorld: Three low-cost, low-fuss VDI solutions prove that desktop virtualization is within anyone's reach. See "InfoWorld review: Desktop virtualization made easy." | Read about the decline and fall of system administration on Paul Venezia's Deep End blog. ]
In order to accurately gauge just how close this race has become, we invited Citrix, Microsoft, Red Hat, and VMware to the Advanced Network Computing Lab at the University of Hawaii, and we put their server virtualization solutions to the test. We compared Citrix XenServer, Microsoft Windows Server 2008 R2 Hyper-V, Red Hat Enterprise Virtualization, and VMware vSphere by virtually every measure, from ease of installation to hypervisor performance, and all of the management capabilities in between.
We tested each solution on the same hardware, with the same real-world network topology, running the same tests on the same virtual machines. We ran real-world and synthetic Linux and Windows performance benchmarks, and we performed subjective management and administration tests. We looked at host configuration, VM templating and cloning, updates and patching, snapshots and backups, and scripting options, and we examined advanced features such as load balancing and high availability.
The results showed that all four solutions combine very good hypervisor performance with rich sets of management tools. But the solutions are not all equal in either performance or management. Although VMware is no longer the only game in town, choosing an alternative certainly involves trade-offs.
VMware still has advanced capabilities that the others lack. VMware also offers a level of consistency and polish that the other solutions don't yet match. The rough edges and quirks in Citrix, Microsoft, and Red Hat aren't showstoppers, but they demonstrate that these alternatives all have hidden costs to go along with their (potentially) lower price tags.
Virtualization shoot-out: The test bed
The fine folks at Dell were kind enough to lend us a bunch of high-end gear to run all of our tests. We requested blade servers for a variety of reasons. Primarily, we wanted the ease of setup and configuration offered by the blade chassis, which consolidates the power, network, and remote management into a single unit. We chose the two-socket blades for our test servers, as these are more representative of production hypervisor configurations than other CPU densities.
We were equipped with a Dell PowerEdge M1000E chassis with two Dell PowerConnect M8024 10G switch modules and a PowerConnect M6220 gigabit switch module. The storage tasks were easily handled by a Dell EqualLogic PS6010XV 10G SAN array, and we used four Dell PowerEdge M710 blades to run the hypervisors. Each M710 was equipped with two Intel Westmere 5645 CPUs running six physical cores at 2.40GHz. Those were accompanied by 96GB of DDR3 RAM, dual-port Intel X-520 10G Ethernet mezzanine adapters, and built-in dual-port gigabit NICs. Each server also had Dell's redundant SD-based flash devices for embedded installations and a pair of 72GB SAS drives in a RAID1 configuration for hypervisors that required traditional installation.
For backline duties, we used two Dell PowerEdge M610 Intel Nehalem-based blades. These blades were not part of the actual test, but were used to provide supporting services such as Microsoft Active Directory, DNS, and DHCP. Suffice it to say, we were very well outfitted on the hardware front.
Next page: What's the world's fastest hypervisor?