I was able to figure out the problem. Per the pdf, 2 esxi hosts are required. I used one to see if I can really push the limits of the hosts but it resulted in this error. I've learned quite a bit working 1 week on VMmark. The pdf is good, but I believe there are 2 bugs with this product that was causing me grief. I'll lay out my experience in hopes that it'll save others from aggravation.
- The biggest issue is name resolution. Vmmark provisions the VM for you and names the VM AuctionWeb0, DSDB0, etc. It spits out the hosts stub for you and you're suppose to copy this to your /etc/hosts file for name resolution. That's great but when running the benchmark, the workload references AuctionWeb , DS-DB, and so forth. They do not reference the VM names with the 0 suffix so I got all sorts of errors. I had to manually modify the hosts file and add additional names of all the VM, less the 0 part and copy the hosts file to every VM. Don't forget about the PrimeClient itself. Make sure it can resolve it's own and from other VMs. The error logs will tell you but you have to search for it.
-2nd problem is the STAF.cfg file. The run process complains about trust level. The default is set at 3. Although I've told it to trust at level 5 for all the VM on a particular subnet, that doesn't work. You have to change the default level from 3 to 5.
-I use distributed switch or dvswitch. I thought the idea was to keep things isolated from production so I created portgroups, ephemeral of course, but I did not bind them to any uplinks so to prevent them from hitting the wire. VMs could ping each other still and VMmark seems to run the tests all the way through but you wouldn't get test scores or you'd get some other errors. This is the same for vmotion portgroups. Make sure it's bound to an uplink so it can touch the wire but doesn't really need to be on the same vlan as production
These were the main stumbling blocks for me and I was able to finally get a test score after resolving all these.
