Lies, Damned Lies and Statistics
The statistics graphed on the article page could benefit from some clarification:
At the moment the graph says that OpenIndiana takes 55 seconds to execute time iozone -a whereas the same task takes O3X almost five minutes. Why is O3X so slow? A likely explanation is that by the time the final VM executes iozone, the virtual disk files are thoroughly cached by the VM host machine and (if applicable) also the SAN/NAS providing them to the VM host machine. Therefore that VM has less I/O wait time. The first VM to run the test has the penalty of warming up the caches (outside the VM) all on its own. There would be two ways to make this fair:
- ensure every VM gets the same penalty by rebooting the VM host (and if applicable the NAS hosting the drive files) between each test to ensure the VM host is in a consistent known state before firing up the next VM; or
- ensure every VM gets the benefit of host caching by looping the test suite twice, and throwing away all the timings from the first loop.
Whatever is the case, it would be good to add some descriptive text to explain why O3X takes 275 sec and all the other implementations take significantly less time.