VMware VSAN: Benchmarking VDI workloads using SanDisk SSDs
Last December, SanDisk® announced our partnership with VMware and Dell for Virtual SAN™ (VSAN). In fact, SanDisk SAS enterprise solid state drives (SSDs), available in Dell PowerEdge® R720 and PowerEdge® T620 servers, have been certified by VMware as a supported flash tier of VMware VSAN. This is important because it allows VMware to provide VSAN customers with support for servers configured with SanDisk SSDs.
VMware Virtual SAN is a new software-defined storage platform that is fully integrated with VMware vSphere®. Virtual SAN aggregates locally attached disks of hosts that are members of a vSphere cluster, to create a distributed shared storage solution. Virtual SAN is a hybrid disk system that leverages both flash-based devices, to provide optimal performance, and hard disk drives, to provide capacity and persistent data storage.
We wanted to help our customers see how they can best leverage VSAN system resources and meet their performance demands. Beyond certifying our products for use within VSAN, we also took part in VMware Virtual SAN™ Beta program and evaluated the product by executing VDI workload on it with SanDisk SSDs. We wanted to share with you our results in this blog.
Our testing used 3-node Virtual SAN built on top of Dell R720 servers. We deployed VMware Horizon View® virtual desktop infrastructure (VDI) environment in this deployment and ran tests using VMware View Planner, VMware’s proprietary VDI workload generator tool. VMware View Planner scoring mechanism was used. This measurement is an indication that desktops sessions are well within the acceptable limit of performance in this deployment. For more information, please refer to VMware VDIMark® Score in View Planner document.
We ran two sets of tests in this deployment:
- VMware Standard VDI Benchmark User Profile (Heavy User Workload) and scaled up to 276 user desktop sessions.
- Custom Power User Profile and scaled up to 300 user desktop sessions.
(Download the full specifications and results from our website)
Our testing results show that Standard VDI Benchmark User Profile, the write latency of the application is zero for most of the time during steady state of the test run with occasional spike (~2-3 times) up to 7-8 msec. It is clear from the below graph that the SanDisk SSD drives in the disk groups are efficient enough to accommodate high volume of user data (high IOPS) with very low to zero latency which resulted in excellent application performance inside each VDI desktops. Take a look at the graph below:
Further, if we look at the test results, it can be seen that CPU usage was at around 95-100%, and prevented further scaling of desktops. That means that if additional CPU resources were available, the storage configuration in this deployment could successfully scale many more desktops.
All these system data will not make sense unless we could manage the application SLA. The below chart shows that the Standard VDI Benchmark User Profile test run well met defined SLA measurement in the three node VSAN environment for disk sensitive operations – below is a CPU and Disk Sensitive Operations Chart (Standard VDI Benchmark User Profile Defined SLA <6 sec.)
For detailed testing results and best practices, please refer to our whitepaper available for download here.
Understanding Disk Groups
One of the key learnings in our experiments is the design of Virtual SAN disk groups. Design of Disk Group (DG) is very important based on the application workload (read, write or mix). Each disk group is a container of SSDs and HDDs. SSDs are used as read cache and write buffer whereas HDDs are used for storage capacity.
As a rule of thumb, if you are executing read intensive workloads, then more SSDs will give better performance since it increases the SSD hit rate. In that case, we need to create more disk groups in order to have better application performance. On the other hand, for write intensive workloads (e.g. full clone, all writes etc.), application data is first written to the SSD write buffer and then flushed to the hard disks in the background. It is very important to balance the SSD/HDD ratio in each disk group so that write buffer of SSDs does not get filled with application data to avoid slow application performance.
As we know, VDI is write-intensive and each user writes lot of data to storage in their day-to-day activities. SSDs (in each disk group) play a major role. If SSDs are not fast enough to push the application data to background HDDs, the write buffer of SSDs will be filled up soon with user data and the application will have to wait to write its data to flash layer. This will result in spiking up disk latency and the end user experience will be severely impacted. Moreover, in Virtual SAN 1.0 version, each disk group can only contain one SSD. So its reliability and consistency is equally important so that it does not fail to avoid adverse impact for such workload.
VDI is a write-intensive workload and application performance in each user session depends on how quickly the data can be written to, or read from, the disk. The lower the disk latency the better the end user experience. The results of our testing, as shared in this blog post, show that SanDisk SSDs are fast and efficient to accommodate demanding VDI disk requirement; high IOPS and low latency ensure successful deployment in Virtual SAN environment, with greater performance and efficiencies.
To learn about high performance VDI using SanDisk SSDs, VMware’s Horizon View and Virtual SAN, please download our latest VSAN Deployment and Technical Considerations Guide.