By Tristan Todd, Architect, End-User Computing, VMware
We have finally brought the plane in for a landing, and it is time to hit the beach!
In this third and final post in a series, we are thrilled to announce that the App Volumes 2.10 Reference Architecture is now available.
Author’s Note: I have always wanted to go to Maho Beach, St. Maarten, and watch the planes land! #bucketlist
Wait, What Are You Doing Again?
If you missed it, review the second post in this series, App Volumes Performance and Scalability Testing – Testing Almost Completed, where we discuss our mid-test findings. Some of the deployment information in the remainder of this post will make more sense if you have read the previous posts in this series.
In the course of this project, we tested App Volumes with over 50,000 View desktops in Horizon 6. We tested a variety of AppStack configurations, collected a mountain of performance metrics, and uncovered some interesting performance characteristics.
User Experience Comparison
After the previous post in this series, we saw an improvement in our environment through further tuning and optimization. Login VSI final VSImax “scores” show that session density declined only slightly after we introduced App Volumes.
Again, it is important to remember that VSImax is the maximum number of active View desktop sessions that our environment could sustain and still deliver excellent user experience. Adding more active sessions would have caused our users to experience a degraded desktop experience.
For a more detailed explanation of VSImax, visit Login VSI’s excellent VSImax FAQ page.
Key Findings
At a high level, we observed that we could operate our 2000-desktop user environment with about 80 percent session concurrency without extreme load on the host servers. Our environment was configured with an 18-host vSphere cluster. A single App Volumes Manager server was utilized. Our hosts and vCenter server handled about 10 percent greater resource load with App Volumes, compared to our tests with native applications. And, finally, with our EMC XtremIO storage, we enjoyed very fast pool operation times. On average, we could deploy a 2000-desktop pool in 80 minutes; recompose operations took a little over 2 hours.
Storage Groups
I fell in love with storage groups during this project. The storage groups feature in App Volumes 2.10 makes it easy to distribute storage resource consumption across multiple datastores. This useful feature also allows you to equally distribute user writable volumes to a group of datastores, providing balance in both performance and capacity consumption. For a fantastic description of this feature that is new with App Volumes 2.10, visit the blog post on VMware App Volumes Storage Group Improvements with 2.10 from Dale Carter of VMware.
In our testing, we used two datastores to host our AppStacks and our writable volumes.
A Few Tips
During this reference architecture project, we developed and confirmed a few operational best practices:
- Administrators must ensure that vCenter servers are built according to current sizing best practices to accommodate the extra 10 percent load that App Volumes introduces.
- Throughout the project, we never observed View desktop virtual disk latency exceeding 5 ms for reads and writes. The all-flash storage that we used helped us deliver fast, consistent performance to our users. Another attribute of the EMC XtremIO storage platform that we enjoyed was the excellent capacity savings through compression and deduplication. We cover this in detail in the reference architecture paper. Customers should strongly consider modern storage systems (ideally all-flash with inline data reduction) for their View desktop environments.
- Customers should always tune the parent desktop images that are used in View. The free OS Optimization Tool is a fantastic way to ensure that your desktop OS will perform well in a View environment.
- We utilized vRealize Operations for Horizon and vRealize Log Insight to help us gather key metrics and study log files during our testing. In a production environment, administrators will find both of these products are very helpful in providing proactive monitoring and reporting.
What Are Your Thoughts?
So, what do you think of the App Volumes 2.10 Reference Architecture? We worked hard to address the most common questions that we have been receiving from customers and partners. We would love your feedback so that we can make our next reference architecture paper even better. Please provide comments to this post and share your ideas!
Special Thank You
Tirtha Bhattacharjee
Girish Narkhede
Stéphane Asselin
Andrew Johnson
Mason Uyeda
Jeff Ulatoski
Sachin Sharma
Jason Marshall
Surya Kiran Cheruvu
Travis Wood
Simon Long
Jim Yanik
Ray Heffer
Dean Flaming
Rick Terlep
Jared Cook
Dale Carter
Todd Dayton
Cameron Fore
Chris White
Tina de Benedictis
Gary Sloane
Dan Dulick
Mark Ewert
Chris Halstead
Yuvraj Mehta
Michael Cooney (EMC)
Chhandomay Mandal (EMC)
Dennis Geerlings (Login VSI)
The project is completed. The plane has landed. Time to hit the beach!
