VMware EUC with Principled Technologies has recently published what is clearly a biased and inaccurate report comparing cost and scalability of XenApp to Horizon View 5.2.
Because some people may read this report and not understand the inaccuracies it contains and the biases behind it, I believe a response is necessary. The community deserves to hear the truth.
It is also important to understand that this blog is not directed towards VMware as a company, rather I wrote this as a response to make sure you–maybe a Citrix and/or VMware customer–see how VMware EUC sponsored reports are spreading misconception and deceiving the community. I consider VMware a great technology company, and I have a lot of respect for many of the technical and business folks at EUC (@Andreleibovici, @DStafford to name a couple).
This less-than-accurate sponsored report also compares non-persistent VDI with RDSH. Here is my blog myth-busting some of those claims. In this post, I want to break down and analyze some of the VMware-sponsored report claims specifically around XenApp vs Horizon 5.2 scalability.
For the purpose of this blog I am going to assume the scalability numbers in the report for Horizon View are accurate. However I do take umbrage at the numbers presented for XenApp and I’d like to start by looking at the claim that XenApp 6.5 can scale to 151 users at lossy compression and 145 users with lossless compression vs 174 users which VMware VDI achieves on identical hardware. The test was conducted using HP G8 hardware with 256 GB RAM and Dell EqualLogic PS-6110X using RAID 10
The XenApp numbers represented in the report are significantly lower than the numbers Frank Anderson (Citrix solution architect) tested in the Citrix lab on a very similar hardware using local storage (no SAN) sometime ago. Frank’s numbers indicate that you can get 203 users sessions with a Login VSI medium workload on Hyper-V2. Is VMware hinting that vSphere has scalability issues to get such poor results? Or are they simply misrepresenting the results by utilizing intentionally poor testing methodology? Note that Frank used an identical VM configuration (4vCPU and 16GB RAM), identical processors (Intel Xeon E5-2680 Processor and used 128 GB RAM which 50% less RAM than this reports configuration). The storage configuration he used featured an HP Smart Array 6 Gb/s PCIe 3.0 SAS controller and eight 10,000-RPM SAS disks set up as RAID 0+1 volumes, but the scalability was constrained by the processor not the storage.
Now, you don’t have to take Citrix’s word regarding the scalability testing, HP has published a non-sponsored and objective XenDesktop reference architecture by HP (see page 23) on similar hardware demonstrating they can get 284 Login VSI light workload XenApp users.
Here is another proof point and another lab report by Citrix Solutions Lab on how to design a XenApp solution to mobilize apps for 500 users. You can get approximately ~205 users using the HP G8 hardware and Intel Xeon E5-2670 Processor CPU, local SSD disks and 50% less RAM than what VMware paper claims. Again, the only difference is that these numbers are on Hyper-V2 (which by the way is free).
In fact, some customers may want to install XenApp on bare metal, which again Frank tested and proved can get ~ 245 users on similar hardware.
The point I am trying to make is that the single server scalability on the identical hardware for XenApp/RDS will be greater than both XenDesktop VDI and VMware View VDI. You can take a look at more alliance partner architectures like Dell DVS reference architecture for XenDesktop and IBM smart cloud desktop infrastructure for XenDesktop
There is no way the numbers for medium workload would be 151 users offering lowering density than VMware VDI, so it’s a shame for there to be so many existing XenApp data points yet such a misleading document sponsored by VMware. There are several other clearly misleading items in this report:
– When taking hardware into the account, the VMware sponsored report conveniently skips the topic of IOPS and the expensive storage requirement for VDI users (Dell Equal Logic PS6110X drive array) for supporting 175 users. XenApp 6.5 can produce high scale numbers even without a SAN or NAS.
– 145 users for XenApp is NOT the Login VSI max in the report, yet the entire report is focused on that number. VMware can you publish the VSIMax for the tests?
– At the 51 minute mark of the View test CPU goes to 100% and flat lines. This does not happen with the XenApp test which means user sessions would be much more responsive for XenApp
– Check out the user response times. For XenApp, all response times are under the 4000ms mark, while this is not true with View.
– There is no mention of total RAM consumed by XenApp. Our experience is and independent reports show that this would be expected to be 50% of what a VDI session would consume
– VMware sponsored report skipped the total amount of storage required for the VDI. The rule of thumb is that each linked clone takes up roughly 15% of the golden image size. Assuming a 40GB golden image, the linked clone storage calculation for 174 pooled desktops would be approximately ~ 1 Terra Byte
If you are really interested in learning about XenApp scalability then I recommend you read the following CTXS article
Bottom line: RDS and VDI scalability numbers are getting closer but there is still at least 15- 20% gap where RDS offers higher user density with no special storage or IOPS requirement and 50% less RAM
In part 2 of my blog I will talk about comparing XenApp cost to VMware Horizon VDI cost.