I am reading the latest review of desktop virtualization in InfoWorld and, as often is the case, Workstation wins.
VMware
Workstation 6.0 solidifies the company’s position as the dominant
player in developer and product support circles. A combination of
class-leading features and excellent scalability make VMware
Workstation the only choice for serious virtualization users.
The author, Randall Kennedy, has benchmarked a beta build of Workstation, which is just silly because we include debug code in our betas that slows them down. He still recommends us, so I don’t want to complain, but c’mon, don’t use beta code for performance evaluations.
However, there’s one correction that just has to be made. The following
paragraph just isn’t right.
As I mentioned above, Parallels is the only
product in this segment to employ a hypervisor — a thin layer of
software that runs below the host OS and provides tighter integration
between the guest OS and the system hardware. It’s a major
architectural advantage for Parallels, one that pays off in the form of
class-leading performance. Parallels outran all comers during multi-VM
benchmark testing …, especially in those tests involving heavy local disk I/O.
When VMware uses the term "hypervisor" — when most people use the term hypervisor — we mean a Type 1 "bare-metal" hypervisor.
This is a hypervisor that sits underneath the operating system on the
bare metal of the server hardware itself. ESX Server is a Type 1
hypervisor, as is Xen. This is as opposed to a Type 2
hypervisor, which runs on top of a "host" OS such as Windows or Linux. Parallels Workstation, like VMware Workstation and VMware
Server, sits on top of a host OS. I’ve been told that Microsoft also
uses the term in some Microsoft Virtual Server collateral. Calling any
of them a hypervisor is just plain confusing to customers and
journalists alike, as is seen in the above paragraph.
Now what Parallels probably means is that their virtual machine
monitor takes advantage of hardware assist (Intel VT and AMD-V) on
newer processors. VMware Workstation also takes advantage of Intel VT
for 64-bit guests, but for 32-bit guests and AMD CPUs, we have a tuned
binary translation (BT) monitor. In fact we’ve shown
that for normal workloads, our BT monitor is as fast or faster than VT,
and therefore we’re more interested in the next generation of these technologies.
So there are small differences in approach between Parallels
Workstation and VMware Workstation, but none that would qualify as a
"major architectural advantage," and indeed in our tests a release build of VMware Workstation performs better than Parallels across the board.
So, be careful about the terminology. Using a word like "hypervisor" is not a magic bullet. VMware Workstation is just as much of a hypervisor as Parallels Workstation is. And when you evaluate software, please use a release build. People who actually need to use this stuff in their daily jobs will thank you for it.
–jtroyer
