World of EUC

We've got a mixed bag of LoginVSI, Enterprise, NVIDA's tooling and Automai (and a bunch of home grown bits for system/function testing)

Oh, just as a disclaimer: When I say we, it's the royal we. I actually mean the team that I try to facilitate to the best of my ability: Performance and Solutions Engineering for EUC at Nutanix :slightly_smiling_face:

ah. that's helpful.  I can see how/why you would use these tools in that role, especially at a vendor.

I'm also wondering what Consultants and Customers are using/doing.

My company is kicking off a POC with LoginVSI. I will say that initially I was not a fan of their proprietary EUC score for determining VSImax, but after thinking about it, I prefer it. VSI does much more now than just performance testing/load testing/capacity planning as well.

What type of testing are you looking for?

There is a product called VDI Drones that can simulate a load on VMs, letting you see how the host behaves. I like it's simplicity, but it leave out the user metric. Since I have to set the specific VM loads for CPU, IO and memory utilization, how do you know exactly where and which will begin to impact user acceptance? With VSI, their proprietary metric will tell you that, and then you can use all the other tools you have (hypervisor console, uber, controlup, etc) to determine what changed from the baseline when performance dropped.

We looked at Lakeside Systrack briefly, but I did not see a component for load testing, just the prod environment monitoring.

Using LoginVSI to simulate user activity and to generate load on a newly build environment to validate the sizing, performance and availability. VSI shows me the data within the sessions (startup times etc). I use Uberagent to get the data from within the session (what proces causes  the CPU load, what part of the login proces is taking to long, what is the latency to a network connection etc). For the hypervisor I use sexigraf to get the CPU Ready, datastore latency's, VSAN performance etc etc while sessions are going up and keep running, etc etc. All there components are relativily fast to deploy and spits out the data without much configuration. I would love to combine the data of the 3 products in one dashboard of even better to get one product to do it all..... but one tool to generate dashboards to relate the data would be realy nice....I know VSI can use splunk, Uberagent uses Splunk,  only sexigraf doens't but I could connect to a splunk datasource to get the data from the other products and display the data from grafana wich sexigraf uses and combine the data..a well..  the time... where is the time...

At GO-EUC we are using primarily LoadGen in combination with own tooling and scripts. Additionally we have access to EUCScore from Benny and depending on the use-case we select the best fitting solution to get to the result. We used to run with Login VSI but due to legal reasons we moved to LoadGen. Quick disclaimer, I'm a former Login VSI employee as well.

In case this is of interest to anyone,  I have created a Slack (freebie) for Login VSI customers.  Happy to have more people join if they want.  It's kind of exploratory now, but hoping to drive more conversation and collaboration around use of Login VSI.

I guess that I should post the Slack invite for the LVSI Slack.  <https://join.slack.com/t/lvsi-customers/shared_invite/zt-1zdagkps2-HCTMBD9g5MKQUZviW3dX6Q>  Feel free to join if you use Login Enterprise or are thinking about using it.

image.png

I think everyone forgets the exclusions.......

But they are massively important and make a huge difference!

yeah. though siloing of sec teams and desktop teams often means security (who is paid to say no) says .... no :slightly_smiling_face:

Yes, and usually the sec teams don't have the time for someone to explain "how it works" so the short answer is no. Security teams need to be more engaged with other teams IMO.

It’s funny when you’re using ‘next gen’ solutions where exclusions are a thing of the past…

I asked this one dude to FLTMC on his machine to look at the number of file system filters. It made my head hurt.

what should happen is for management to see how much its costing them and they can decided if its worth the extra hardware/resources to implement the "no's" or if they should evaluate actual requirements

I think what’s crazy around this these days is that it’s not just agents that can have exclusions configured like AV. The amount of different agents, file watchers, behaviour trackers, software auditors, log forwarding, vulnerability scanners that are in the mix is chaos. Definitely a problem that gets exponentially worse the more you add. 

Found a customer this week who can get 4 users per VM due to agents, something like 8-9 without them. AV rules look like swiss cheese trying to keep up with it all 

Would love to see the additional pressure on the hypervisor &amp; storage... EUX and VSImax is dropping can imagine this is due the extra CPU cycles or IOPS on storage.

For those interested, in about an hour Blair Parkhill (or director on the product side) is going be on with <@ULKDZHKFY> doing a webinar using our Teams 2.0 workloads and looking at Perf impact and some of the testing they have done.  <https://www.brighttalk.com/webcast/19518/608690?utm_source=brighttalk-portal&amp;utm_medium=web&amp;utm_campaign=channel-page&amp;utm_content=upcoming>

Hi all.
We're setting up an AVD pool on Azure Stack HCI for a client migrating from VMware Horizon and we want to show the performance from both solutions.
I'm using diskspd for storage testing but what should be used to test CPU, memory and perhaps GPU?

Login Enterprise, Dynamik Loadgen or AutoMai. 

Diskspd might give specific results depending on the platform you’re testing 

run load up. with your image, your security package, your authentication, etc, etc, etc.

Once you ca do that you push that real idea of what you need to support the requirements.

Anders, if you want a demo of Login Enterprise and what it can do, DM me and I'll be happy to oblige. Happy to feed in with my experience in the field, performance testing solutions like this also.

Defender Comparison.pdf

Want to see what happens to your density and performance in granular detail when you don't have A/V exclusions in place?

Check out this report produced in the latest release of Login Enterprise.

Great test <@ULKDZHKFY> -  it's not clear from reading the PDF that the comparison is with/without DFE configured...also an appendix would be useful of what Defender configuration you put in place vs the vanilla baseline

His next phase he is putting together perhaps? He is really good about getting all those nuggets in time

Good points gents, this was actually a new feature in the product to just highlight two tests and produce a report. Next release gets more customisation to allow for removal/addition of different sections. You think a block of free text would be good so you can highlight these things.

Neil, you're right, testing methodology is missing.

For reference though, Vanilla is just OOB Defender, Optimised is only Citrix exclusions and scanning on writes only (Non-Persistent Environment).

If I hadn't read the text here, I wouldn't know what comparison was being made, just from the PDF

<@U05U79WD0R1> - Feedback I'll take to product.

Neil, Ray - I'll sort you out a T-Shirt.

(It doesn't matter if you already have one Neil :joy:)

Is there a presumption that the gold image has been scanned to turn off read-scanning?

Yes indeed Neil, that is the case. BIS-F handling this on sealing.

Feedback has been given to product and it will end up in the product with regards to the "Load test free text intro" of the PDF report.

Have a client with NC8AS_T4 backing a old Server 2016 VDA. Publishing a full desktop those max out at around 10-12 users generally, or 5 if they all decide to do heavy GIS workloads at same time like in training sessions.

hey gang. i wanted to get a feel from the rest of the community as to what you're seeing for user density on your multi-session VDAs. doing Win10 or Server 2022, we're hitting a CPU bottleneck with around 12-15 users. These are Azure NC_AS series VMs (16 vCPU with a GPU and using ephemeral disk. I would hope that we can get close to that 20 user mark and I believe there may still be some tweaking needed to get more. These VMs are 16 vCPU and 110gb ram so i'm not getting a user per cpu :disappointed: you guys mind sharing what you're seeing?

I assume this is because of the GPU; in general, when adding a GPU, you will see a bigger hit on the CPU, which is definitely going to affect the density. I have never tested this SKU, so I don't have any performance results. Maybe <@ULCLL3M41> has some...

I don't have performance data with this SKU and MultiUser :disappointed:  - that being said, can you share more applications being used?

I think <@ULHA2J7RT> added some anecdotes on N series last time we had a A, B, D, E, N instance of choice type thread....

You might also get more replies on the general channel

We use the NC16as_T4_v3 series VM for CAD users. Therefore our limit is VRAM.
During our load tests with AutoCAD we calculated a max user density of 4 users, which works very well.
So in fact I have data, but for a completely different use case.

I haven’t done much testing without a gpu but I was under the impression that a gpu would actually lessen the load on the cpu

I would say the only answer we can give you is: It depends on the actual workload.

Older post, but these were the findings back in the day. I expect it's still the same with modern GPU's
<https://www.logitblog.com/conflicted-results-with-nvidia-tesla-m10/|Logit Blog - Conflicted results with NVIDIA Tesla M10>

We're in the banking industry and lean heavily on the whole 365 suite. 365 will supposedly utilize a GPU and Teams garbage as well. Nothing real heavy from that aspect. makes me wonder if i can change studio policies so it doesn't leverage the GPU and do some testing on the same VM SKU

You can disable those apps from using GPU with a  GPO setting for Office and a json setting for Teams.

You’re not going to get more than 15 users these days no matter how big the VM unless you’re just doing apps. 

There’s almost no benefit past 4cpu and apps will eat up as much ram as you give them

We see 4cpu 32gb ram and capping at about 10 users to be the sweet spot 

Any vm. 15 users is basically it. GPU or not. 

I haven’t found any combination of hardware that will make the user experience better when there’s more than 15 users on a multi session VDA 

Crazy to think we went from 100+ users on bare metal servers back in the day to only 15-20 users now!

Yeah that’s my point. I remember getting +/- 50 on some old shitty dell blades

you still can! Just go back to Server 2003.

Bild von iOS.jpg

lol, this just goes to show how much people hate linux

Oh please don’t misunderstand me. That was regarding your argument that 2003 had a much better user density. That was during my training for my job where we had Citrix on 2003 and performance was great I guess. 
But for example our Zabbix Monitoring for all of our MSP customers runs on Debian administered by me. 

yeah I meant more for endpoints, like we saw with crowdstrike people would rather put windows on a bidet than have to deal with embedded linux