VMware vCenter Server™ 6.0 substantially improves performance over previous vCenter Server versions. This paper demonstrates the improved performance in vCenter Server 6.0 compared to vCenter Server 5.5, and shows that vCenter Server with the embedded vPostgres database now performs as well as vCenter Server with an external database, even at vCenter Server’s scale limits. This paper also discusses factors that affect vCenter Server performance and provides best practices for vCenter Server performance.
What’s New in vCenter Server 6.0
vCenter Server 6.0 brings extensive improvements in performance and scalability over vCenter Server 5.5:
- Operational throughput is over 100% higher, and certain operations are over 80% faster.
- VMware vCenter Server™ Appliance™ now has the same scale limits as vCenter Server on Windows with an external database: 1,000 ESXi hosts, 10,000 powered-on virtual machines, and 15,000 registered virtual machines.
- VMware vSphere® Web Client performance has improved, with certain pages over 90% faster.
In addition, vCenter Server 6.0 provides new deployment options:
- Both vCenter Server on Windows and VMware vCenter Server Appliance provide an embedded vPostgres database as an alternative to an external database. (vPostgres replaces the SQL Server Express option that was available in previous vCenter versions.)
- The embedded vPostgres database supports vCenter’s full scale limits when used with the vCenter Server Appliance.
Performance Comparison with vCenter Server 5.5
In order to demonstrate and quantify performance improvements in vCenter Server 6.0, this section compares 6.0 and 5.5 performance at several inventory and workload sizes. In addition, this section compares vCenter Server 6.0 on Windows to the vCenter Server Appliance at different inventory sizes, to highlight the larger scale limits in the Appliance in vCenter 6.0. Finally, this section illustrates the performance gained by provisioning vCenter with additional resources.
The workload for this comparison uses vSphere Web Services API clients to simulate a self-service cloud environment with a large amount of virtual machine “churn” (that is, frequently creating, deleting, and reconfiguring virtual machines). Each client repeatedly issues a series of inventory management and provisioning operations to vCenter Server. Table 1 lists the operations performed in this workload. The operations listed here were chosen from a sampling of representative customer data. Also, the inventories in this experiment used vCenter features including DRS, High Availability, and vSphere Distributed Switch. (See Appendix A for precise details on inventory configuration.)
Figure 3 shows vCenter Server operation throughput (in operations per minute) for the heaviest workload for each inventory size. Performance has improved considerably at all sizes. For example, for the large inventory setup (Figure 3, right), operational throughput has increased from just over 600 operations per minute in vCenter Server 5.5 to over 1,200 operations per minute in vCenter Server 6.0 for Windows: an improvement of over 100%.
The other inventory sizes show similar gains in operational throughput.
Figure 3. vCenter throughput at several inventory sizes, with heavy workload (higher is better). Throughput has increased at all inventory sizes in vCenter Server 6.0.
Figure 4 shows median latency across all operations in the heaviest workload for each inventory size. Just as with operational throughput in Figure 3, latency has improved at all inventory sizes. For example, for the large inventory setup (Figure 4, right), median operational latency has decreased from 19.4 seconds in vCenter Server 5.5 to 4.0 seconds in vCenter Server Appliance 6.0: a decrease of about 80%. The other inventory sizes also show large decreases in operational latency.
Figure 4. vCenter Server median latency at several inventory sizes, with heavy workload (lower is better). Latency has decreased at all inventory sizes in vCenter 6.0.
Download a full VMware vCenter Server 6.0 Performance and Best Practices Technical White Paper