I’m a few weeks away from publishing the Reference Architecture for Horizon 7 with App Volumes on All-Flash Virtual SAN but I just wanted to give our customers and partners a preview of our test results.
Here are 3 key takeaways for anyone managing or planning a VDI environment:
- Just-in-time delivery with Instant Clone Technology: This first-to-market capability from VMware provides a new and dramatically accelerated means to provision fully featured, personalized, and customized virtual desktops.
- Further improved TCO with Virtual SAN 6.2 storage efficiency features such as deduplication and compression, erasure coding, and sparse swap.
- Virtual SAN provides the simplicity of managing storage performance and capacity in a single, tightly integrated interface–the vSphere Web Client.
In Virtual SAN 6.2, we introduced key space efficiency features such as deduplication and compression, and erasure coding (RAID 5/6) to reduce capacity consumption while ensuring the same levels of availability and performance for a lower total cost of ownership. During our testing, we used Login VSI 4.1 Knowledge Worker workload benchmark mode to measure VDI performance in terms of VSIMax score. We used an 8-node All-Flash Cluster with 1,000 virtual desktops, and a management cluster consisting of a 4-node hybrid Virtual SAN with the Login VSI launchers. Both instant clones and linked clones achieved over 120 VMs density per host for the knowledge worker workload. That means that with all of the space efficiency features of Virtual SAN 6.2 enabled, Virtual SAN provides great performance with minimal impact.
All-Flash Virtual SAN Specifications and Performance
Server Specification (per Host)
- 2 x 10 Intel(R) Xeon(R) CPU E5-2690 @ 3.0GHz v2
- 512GB RAM
- SSD: 2 x 800GB Solid State Drive (Intel SSDSC2BA40) as Cache SSD
- SSD: 8 x 400GB Solid State Drive (Intel SSDSC2BA40) as Capacity SSD
Virtual Desktop Template Configuration (VMware OS Optimization Tool was used to optimize the image)
- Windows 7-64bit
- 2GB memory
- Disk size: 30GB, 10.6GB used
- One AppStack of 6,536MB was created based on the default AppStack template. The applications included Adobe_Flash_Player_16_ActiveX, Adobe_Reader_XI_-11.0.10, Doro_1.82, FreeMind, Microsoft Office Professional Plus 2010. All applications from the AppStack along with default Windows were used in conjunction for LoginVSI testing.
We used Login VSI 4.1 to load the system with simulated desktop workloads using common applications like Microsoft Office, Internet Explorer, and Adobe Reader, and so on. The VDI workload in general can be CPU intensive. Virtual SAN can support up to 200 desktops per host from the storage perspective if the host CPU is sized properly. During the Login VSI testing, we found that our servers were CPU bound. Therefore, we focused our tests on 1,000 virtual desktops.
First we focused on the Login VSI benchmark on instant clone pool with the default RAID-1(Mirroring) policy. Then, we changed the policy to RAID 5 (Failure tolerance method= RAID 5/6 (Erasure Coding-Capacity). VSImax knowledge worker v4.1 was not reached in both configurations. The instant clone testing both passed the score of 971. For policy RAID 5, it has minimal impact on application response time.
Then we focused on Login VSI testing on linked clone pool for various configurations as shown in the chart below.
Virtual SAN 6.2 has enhancements on caching, which adds a small in-memory read cache at the client side. Small in this case means 0.4% of a host’s memory capacity up to a max of 1GB. It is enabled by default. We compared the performance with the client cache disabled.
VSImax knowledge worker v4.1 was not reached in any of the test configurations. Client cache improves the application response time and the new Virtual SAN 6.2 space efficiency features have little performance impact in terms of VSIMax score. The linked clone baseline testing passed the score of 971; the linked clone deduplication and compression+RAID5+checksum testing passed the score of 971; the linked clone deduplication and compression+RAID5+checksum+client cache disabled testing passed the score of 967. We can see from the chart below that the response time was much lower when client cache was enabled.
Space Saving by Enabling Deduplication and Compression and EC (RAID5) Policy and Sparse Swap Files
We measured the space savings of a real VDI environment deployment, provisioned the linked clone pool on an All-Flash Virtual SAN Cluster with deduplication and compression, and erasure coding enabled. In addition, we used the Virtual SAN sparse swap files, which enables “SwapThickProvisionDisabled” advanced host setting for space savings on both linked clone pool and instant clone pool.
Deduplication and compression are applied on a “per disk group” basis. The results of deduplication will vary for different kinds of data. Erasure coding guarantees storage resource savings without compromising on data availability.
Virtual swap files are sized according to the allocated memory minus reserved memory. In large deployments of thousands of virtual machines, this additional capacity could be substantial. By enabling sparse swap files, the swap file will be thin provisioned and disk space will only be claimed when the swap file is consumed. We only recommend using this when you are not overcommitting on memory.
The template VM disk size is 30GB. We provisioned a dedicated linked-clone pool with 1,600 virtual desktops under the policy of RAID 1 (mirroring) and software checksum was disabled, the total space usage was 13.59TB. After enabling deduplication and compression, we changed the SPBM policy to RAID 5 (erasure coding) and the space usage decreased to 7.94TB. The deduplication and compression ratio was approximately 1.91x. Since our servers had enough memory, we powered off all the desktops and configured the sparse swap files; then we powered on the desktops, the corresponding used space decreased to 1.81TB and the deduplication and compression ratio increased to 11.60x.
We also compared sparse swap and SPBM policy RAID 5 on instant clone pool with 1,600 VMs. When sparse swap was enabled, 1,600 instant clones saved 6.27TB with RAID 1. Compared to SPBM RAID 1 setting, policy RAID 5 setting saved 0.68TB.
In addition to running Login VSI benchmarking tests, VMware View operations tests were conducted to measure the effect of these administrative tasks on the environment. Instant clones dramatically accelerated provisioning for fully featured customized virtual desktops. With default setting “Max concurrent Instant Clone Engine provisioning operations=20”, it took only 37 minutes in total for 1,600 desktops to become “Available” with policy RAID 1. And instant clones do not need to be refreshed, recomposed, or rebalanced. When a user logs out of the desktop, the desktop always gets deleted and recreated as a fresh image from the latest patch. This process creates a staggered approach to patching, thus eliminating “boot storms,” which only took 38 minutes to update to a new image. It took similar time with policy RAID 5.
For linked clone operational tests, it took a little bit longer time for Horizon operational execution when you enable space efficiency features, but it is worthwhile for storage capacity saving. The executed time was measured with setting”Max concurrent View Composer provisioning operations=24”.
Virtual SAN is optimized for modern all-flash storage with space saving features that lower TCO while delivering incredible performance. Virtual SAN 6.2 is ready for VDI with tested and validated deployments of Horizon 7 combined with App Volumes 2.11. This blog is a preview of a comprehensive reference architecture paper that is being published very soon, stay tuned.