Dell PowerEdge R820 - Manual (Page 15)

Dell Virtualization Solution for Microsoft SQL Server 2012 using PowerEdge R820
15
Impact of Hyper-threading on Large VMs Figure 8.
Until 4 VMs, the performance numbers are similar between the HT enabled and disabled configurations.
Beyond 4 large VMs, Figure 8 shows that the batches per sec delivered per VM are better when we have
HT enabled in the system. With 7 Large VMs, the server processor utilization was around 86.1% and 98%,
on HT enabled and disabled respectively.
Visit http://msdn.microsoft.com/en-us/library/cc768535%28v=bts.10%29.aspx for more information on
measuring Hyper-V performance.
The Hyper-V counters collected during SQL workload tests are given in Appendix A. It illustrates the
difference in the Hyper-V performance counter while running the VM1 to VM7 concurrently.
Conclusion
Dell PowerEdge R820 is a powerful server platform to accommodate a virtual environment due to its
large memory support, I/O options and high processing power. The support for the latest Intel® Xeon®
processor E5-4600 product family and the high speed(1600Mhz) memory options make it a high
performing platform too.
The experiments carried out as part of this white paper showed that the virtual machines scaled up
without any major performance impact until up to 85% to 90% of the overall server utilization. Using a
4 processor cores per socket (HT disabled), a single R820 was able to scale up to 4 large virtual
machines (each with 4 vCPUs). At the same time, it could accommodate 7 Medium virtual machines
(each with 2 vCPUs) without any major impact in the performance.
We observed a performance degradation of individual VMs once the entire server utilization was
beyond around 90%. This observation makes it obvious to have a constant monitoring of the overall
utilization of the virtualized system and sizing it optimally to meet the workload requirements.
Extrapolating the scalability test results (HT disabled), we can summarize the database consolidation
capability of PowerEdge R820 for the specific OLTP workload as given in Table 3.