VMware Horizon 6 with View – Hosted Shared Virtual Desktops with Nutanix

With Horizon 6 adding support for RDS, Application pools is getting a lot of buzz. With application pools, you can deliver a single application to many users. The application runs on a farm of RDS hosts. However, you can use a farm of hosts to deliver hosted shared desktops (HSVD). I suspect in reality this will shift lots of workload over at least from the task worker use case.

To get HSVD setup with Horizon 6 you have to:

Once the farm is created you can go to desktop pools and pick the appropriate option.

Nutanix Value

1) Quick Clones\VMCaliber Clones – Horizon 6 does not support View Composer for RDS so there could be a potential for lots of storage to be gobbled up. VMCaliber Clones have no negative impact on performance, allow for fast deployment and is available in every Nutanix software edition.

Check out the space saving from quick clones. 84 GB to 12KB

Check out the space saving from quick clones. 84 GB to 12KB

2) Data Locality & Fair Share – Fair Share from 2012 R2 to gives predictable user experience so one user does not negatively impact the performance of another user’s session. Combined with data locality as the cluster scales, IO performance will be consistent and not users can’t steal or bleed the rest of the performance from users on other nodes.

3) Tunable Redundancy Factor – Starting with the Pro Software Edition and up you can allow VM’s to have greater resiliency by creating additional copies of data. Since VMCaliber clones(per vm snaphots\clones) are reducing the foot print the added capacity cost of higher Replication Factor will be mitigated. Now you can lose up to 2 nodes on a 5 node cluster as an example without having to buy additional HDD’s for capacity. Additional block awareness can let you lose an entire block (4 nodes/servers) at the same time without downtime, without requiring any extra space! This all adds up to more capacity for other server workloads in your environment.

Hope see you at VMworld 2014 and talk more on this topic.



EUC Tip 94: Dedupe your base image with no performance overhead. #Xen #View #RDS

Nutanix uses a scaling approach for sizing the Controller Virtual Machine (CVM) that lives on each node/server in the cluster. 8 GB of RAM is reserved for services like Zeus(config data), Medusa(Metadata), Prism(UI), Cluster Health and Cerebro (DR). Pretty efficient considering all the enterprise features the CVM is delivering. The remaining RAM is used for Stargate which is responsible for servicing IO requests. Of the remaining RAM 40% is used for the extent cache and content cache (Deduped cache).

How Nutanix scales the local cache for dedupe and non-deduped data in NOS 3.5

How Nutanix scales the local cache for dedupe and non-deduped data in NOS 3.5

Stargate RAM = CVM RAM – 8G (typically)
Total Available For Cache = 40% of Stargate RAM
Extent Cache = 60% of Total Cache (typically; but we give at least 4G when possible).
Content Cache = Total Cache – Extent Cache (but min of 1G; otherwise off)

You can see when sizing the CVM between 20 – 24 GB the extent cache stays static. If you have a workload that not using VAAI like using MCS, view composer(non-VCAI) or RDS with the applications built into the base image you can take advantage of the content cache without any overhead of inline dedupe. My simply doing a manual fingerprint(SHA1) of the golden image you can utilize this space. This will save lots of space in the content cache and ensure the most used bits never leave your overall cache even in the advent of a really nasty anti-virus problem.

To manual fingerprint the base image you can log into anyone of the CVM’s and run this command:

nutanix@cvm$ vdisk_manipulator --operation="add_fingerprints" --nfs_container_name="ctr_name" --relative_file_path="vdisk_path"

Replace ctr_name with the name of the container where the vDisk to fingerprint resides.
Replace vdisk_path with the path of the vDisk to fingerprint relative to the container path (for example, Win7-desktop/Win7-desktop.vmdk).


EUC TIP 91: Best Practices with Snapshots with RDS

All reference VMs must be generalized before they can be used to provision or recreate a virtualized desktop

As best practice, start with a snapshot of an ungeneralized Windows 7 or Windows 8 virtual machine

Generalize, shutdown, and create new snapshot
In the virtual machine, run the following command from Windows\System32\Sysprep directory
Sysprep /oobe /generalize /mode: vm

The switch Mode:vm only detects the enumerated and synthetic devices that are possible and skips on the physical layer checks. Only works with Windows 8 and 2012

The ‘now’ snapshot reflecting the generalized state will be used to (re)create the virtual desktops

When updating a VM, revert to ungeneralized snapshot, delete generalized snapshot, update VM, generalize, shutdown. and snapshot

I think this process can eliminate the need for a KMS server with RDS but not 100% sure, let me know if you have the answer.


EUC TIP 89: Embarrassingly Parallel Processing for Microsoft RDS

Hyper-V 2012 defaults to deploying VM’s one at time, slow and steady. If you think about it, that can be a lot if you have a 64 host cluster you are deploying desktops too. Nutanix runs a virtual storage controller on each node/server so we handle a little bit of blunt force trauma without wiping out the rest of the workloads. You can refer to my mixed workload post to get the specifics.
[Read more…]


Hyper-V and VDI on Nutanix: Part 1

Screen Shot 2013-12-13 at 10.03.10 AM

If you’re a regular visitor to the blog you know that I have lived in the lands of VMware VDM, View, Horizon and probably whatever the next name is. I wanted to deploy Remote Desktop Services on Hyper-V since requests from the field has started to come since we announced support for Hyper-V.

You still get the same quick time to value as you did from vSphere deployments on Nutanix, not a lot of changes it that department. Nutanix engineers did add some deployments scripts that automate adding the cluster to the domain and to SCCVM if present.
[Read more…]