Hi Daniel,
We're glad that you are interested in using VMmark for academic purposes.
A official run of Vmmark (i.e. one that can be published on the VMmark website) is made up 3 application workloads over a minimum of eight different virtual machines running concurrently on an ESX host cluster. Additionally, VMmark creates periodic vMotion, Storage vMotion and VM Deploy operations that stress the infrasturcture. So, the VMmark test is not executed in one virtual machine, but made up of at least one set of eight virtual machines plus infrastructure operations.
However, the rules for using VMmark in an academic paper or research differ from those for publishing an official benchmark result.
For example, if you are using VMmark for academic purposes, you are not required to run all workloads. You could choose to enable only one, some, or all workloads at a time. This means that it is likely you can run at least some VMmark workloads on an ESX host that has inferior hardware to what's recommended, because the recommendation assumes that all workloads will be running. I'd recommend starting with only one workload running, then add workloads incrementally to be sure your host can handle the load.
You mention that you have only one ESX host. An official run requires two hosts to support the vMotion operation. However, if you have only one ESX host, you should disable the vMotion operation in the VMMARK2.CONFIG file. You would also need to set DoClusterMapping=0 (in the VMMARK2.CONFIG file to avoid errors. All other workloads should be fine running on only one host.
In addition to your ESX host and vCenter machine, you will also need a minimum of one other physical machine installed with Windows Server 2003 or 2008 to serve as the benchmark's client.
VMmark's rules for academic and research use are outlined in the VMmark Run and Reporting Rules; be sure to give these a careful read-though. For example, an academic paper should not attempt to make competitive comparisons using VMmark results, and should disclose any way in which the benchmark differs from its official use, e.g. if some workloads have been disabled.
Please feel free to post back here if you have any further questions.
Rebecca