How cool is your virtual (Xen)desktop?

Hot or not?

If you’ve recently taken a look at the very comprehensive Citrix Validated Solution for Nutanix you must have noticed the high density footprint you can achieve when running XenDesktop on our hyper-converged solution. It’s not just a drastic simplification of the datacenter infrastructure needed to run a desktop virtualisation environment on, it also makes it a hell of a lot smaller in physical appearance as well.

One thing that’s not covered specifically in the CVS is what this actually means for the running costs of an environment like this.

Stating the obvious would be that as part of building any business case there should be a proper CAPEX/OPEX analysis and to show you the impact of running costs I will compare our solution with three competitor’s reference designs for XenDesktop that are publicly available.

A couple of assumptions and ground rules before I start:

  1. The Nutanix footprint is based on 110 VDI VM’s per node, or 200 HSD (i.e. XenApp)  users per node
  2. The first two Nutanix nodes include the infrastructure VM’s as shown in the diagrams below and are capable of also running an amount of desktops as well. The first diagram shows the infra on the first block for HVD’s and the second one shows the infra for the first VDI block.InfraHSD InfraVDI
  3. Performance across the different vendors is assumed to be equal as all designs have been tested with the LoginVSI Medium Workload or other benchmarks to show performance is good.
  4. Cost per kWh is set to the Dutch industry average of EUR 0.23 per kWh
  5. Power requirements and BTU are taken from the vendor’s specification sheets, or tools like the online Cisco UCS power calculator and are the typical values, or reduced to 75% load if only max was given. Full list of used materials is available at the end of this post.
  6. Cost per desktop is calculated over the total infra, as also brokers & switches etc are needed to run to make a desktop available.
  7. Calculations for cooling where made as described in this blog posting by Edwin Weijdema.
  8. XenDesktop versions differ slightly (7.1 and 7.5) , as do hypervisors used. We use Hyper-V while the other designs use VMware.

Nutanix vs VCE vBLOCK

With the ground rules established and the playing field levelled let’s show you the first footprint comparison;  a VCE vBLOCK vs Nutanix for a 100% XenDesktop VDI environment of 2500 desktops. You can download the VCE design here.

Let’s take a look at the rack footprint first.

RackvBlock

The vBlock nearly takes a full rack (35U out of 42U) to accommodate those 2500 desktops, whereas the Nutanix environment uses 20U. In total we would need 25 nodes with the top block just housing one node. 6U’s of our rack is taken by the Arista TOR switching infrastructure.

Because the vBlock design is based on using Provisioning Services and it runs 4 PVS servers, I’ve removed the desktops on the first two nodes, to accommodate for the 4 extra PVS servers. This means the 2500 desktops themselves would run on 23 nodes (2530 total desktops).

Now lets take a look at the numbers:

MatrixVblock1When running a configuration like this, there is obviously not just the cost of the power, but also the cost of cooling the generated heat. According to Edwin’s blog for every 1000 BTU/hr you generate, you need 293 Watts/hr of energy to cool it back down.

This means a vBlock will cost you a staggering total of EUR 69547 per year to run and cool, based on 75% load. Compare this to the Nutanix configuration which uses just half of that. Calculate this back down per desktop and it will cost you EUR 27,81 per year to run a desktop on a vBlock, and just EUR 13,85 to run it on Nutanix. Overall cost savings are up to 50%. How do you like them apples eh!

Nutanix vs Flexpod.

The Cisco / NetApp reference design, which can be found here, is made for a total of 2000 users, 75% being HSD and 25% being VDI.

The rack footprint:

RackFlexpod

Again huge difference in backspace needed; 30U vs 11U. A total of 4 blocks are needed for Nutanix, with the top block filled with 2 nodes. The two infra nodes supply 100 HSD users each, so we need about 9 nodes total to house the 1500 HSD users, and another 5 to house the VDI users. We actually overshoot by 100 HSD users, and 50 VDI users, so you could more or less take a node away but let’s leave it like this for sake of simple calculations.

The numbers:

MatrixFlexpod1The numbers speak for themselves. 60% running costs savings per year, with a per desktop cost of just EUR 9,49 per year. Getting sub EUR 10,- running costs per desktop is off course also a credit to using the Hosted Shared Desktop Flexcast model, which has a much higher user density compared to the VDI footprint, but its obvious less equipment means less cost.

Nutanix vs Cisco + EMC VSPEX

Finally lets take a look at the Cisco + EMC VSPEX design. This design is for running 1000 users with a 70%/30% split between HSD and VDI. Our footprint is pretty much half of the footprint compared to the Flexpod footprint, but we don’t see much reduction with the Cisco + EMC VSPEX combo which proves how inefficient legacy storage is for running large desktop virtualisation workloads:

RackVspex

Based on our CVS we need a total of just 7 nodes to spread the 750 HSD and 250 VDI users, which leaves us with 4U for the appliances and 3U for the TOR switches.

Here’s the numbers:

MatrixVspex170% costs saved, or EUR 25654 per year, with a cost per desktop of EUR 11,23 per year. Rackspace saving of a staggering 80%.

Conclusion

It’s not a big surprise that a hyper-converged infrastructure will give you a drastically smaller footprint for your Citrix environment (or any other workload for that matter). Not only do you need less components and less amount of RU’s in your rack with a Nutanix infrastructure; powering those legacy components and cooling them costs a fortune each year. A lot of legacy deployments require the installation of more or extra power lines into the datacenter including all the equipment that goes with it. Sometimes this might not even be possible at a location.

Less is more.

 

List of used sources:

 

7 thoughts on “How cool is your virtual (Xen)desktop?

  1. Updated the article because of a small calculation error in the Excel I used.

    Net result is that the advantage swings even more towards Nutanix…

    • Hi Francis,

      Thanks for your effort, much appreciated!
      Make sure to update the tables though, because I corrected some small calculation errors.

  2. Martijn, awesome post! And these scenarios are just the beginning! Because of Nutanix’s ability to scale out linearly, you 1) don’t have to stop here, and 2) do not have to make disruptive or forklift upgrades!

    I need to recalculate these numbers for the US so I can start using them over here. 😉

Leave a Reply

Your email address will not be published. Required fields are marked *

*