Recently, SimpliVity released its reference architecture (RA) for VDI. The RA includes an amazing benchmark for the number of VDI users that can be hosted on each OmniCube node. The benchmark results show acceptable desktop performance with significantly more users per node compared to other results.
The natural first reaction to these density numbers is to question the validity of the benchmarking methodology. However, SimpliVity’s benchmark process has been validated by LoginVSI, which created the benchmark software. The LoginVSI benchmark is the standard for capacity testing VDI platforms. It is a real-world benchmark. This means that the benchmark software drives normal applications to generate load. Synthetic benchmarks directly create load on the VMs. Real-world benchmarks more closely resemble real users doing their daily work. To be a useful benchmark, the applications are driven by some automation, providing a consistent workload.
Benchmark results are no guarantee of how a particular platform will respond to your users—this isn’t their purpose. They do allow you to compare different platforms as you are selecting a platform for your VDI deployment. LoginVSI offers services to validate that their benchmark software has been used correctly and that customers can trust the results. SimpliVity used this service to validate that their testing followed sound methodologies and produced a reliable result.
SimpliVity built out a reference architecture for 2,000 user desktops in office worker configuration using VMware View. The user desktops were Windows 7 with Microsoft Office 2010 and a small collection of supporting applications.
This is a slightly higher load than the task workers that are commonly used for VDI hero numbers. Task workers reflect a call center-like environment while, office worker is more like a head office environment. This workload type more closely resembles widespread VDI use within an organization and as such is more helpful in planning a VDI platform. The 2,000 desktops were placed on a total of 10 SimpliVity nodes—two clusters of five nodes, for an average of 200 user desktops per node. Although the benchmarks were run with significantly less than the recommended RAM in each node, they still showed excellent responsiveness for users. Even when one of the nodes was intentionally failed, the desktop responsiveness was good.
The SimpliVity RA is also nicely loaded with details of how the benchmark was run, including the View and Windows infrastructure as well as all the optimizations. The ESXi server settings that are not the defaults are also documented, as are the optimizations that were applied to the Windows desktops. This sort of openness allows customers to have confidence that their deployment reflects a tested best-practice deployment of VDI on the SimpliVity platform.
The SimpliVity hyperconverged platform is well suited to VDI workloads. Having large numbers of VMs created from the same base image means that SimpliVity’s data efficiency enables great storage performance. By deduplicating the VM storage, SimpliVity is able to get maximum performance out of a relatively small amount of flash in each node. The ability to host 200 office worker desktops on each SimpliVity node is a great proof point for deploying VDI on SimpliVity.