One of the most expected solution at upcoming VMworld 2006 is VMmark: a reliable benchmarking system to measure virtual machines performances.
After gradually introducing the benchmarks topic during these months, VMware finally reveals some details about this system with a new 14-pages whitepaper:
Clearly, a more sophisticated approach is required to quantify a virtualization environment’s performance and ability to run an increasing number of diverse virtual machines as physical resources increase.
First, all relevant hardware subsystems should be exercised as they would in an actual datacenter. The individual virtual machines should also operate at less than full utilization to mimic consolidation within a datacenter environment. The benchmark must scale in a controlled fashion to make comparisons between systems meaningful. Small fluctuations in the performance of individual virtual machines can be used to discern minor differences between similar systems. Larger gaps in performance can be measured by increasing the number of active virtual machines. The benchmark must also exhibit stable, reproducible performance.
This paper presents a benchmark, VMmark, to address these goals. The paper is structured as follows:
- Workload Tiling on page 2 introduces the concept of a multi-workload tile that encapsulates several diverse workloads.
- Workloads on page 3 describes the individual workloads
- The scoring algorithm is presented in Scoring Methodology on page 6
- Scores from a 2-CPU server are presented in Experimental Results on page 7
Read the whole paper at source.