Normal UEM login times

[sjesse]

What sort of login image do we expect with UEM to add with just folder redirection and importing Archives on logon. I created a new rds server that I clock a horizon published app at launching within 5 seconds, this is with uem and appvolumes not involved. Adding UEM added 6 seconds(total of 11) and no condition sets are directly targeted to this user or machine other than folder redirection.

I have new customer I'm working with and I need to make sure everything is optimal as possible, so I was wondering whats the best performance people have seen. I have a new all flash "hci" enviornment so hardware-wise I don't think I can get much better.

[DEMdev]

Hi sjesse,

I'll leave it to people who actually use the product to provide their figures , but can you share a DEM/UEM log file so we can see what exactly is going on during those additional 6 seconds?

[sjesse]

Sure, I just santized the domain names

[valkyr0]

We have a fairly large DEM (UEM) environment with ~27K users. Most app profiles are sub-1MB. There are ~50 conditions to go through, mostly shortcuts but a few ADMX configs. We have it hosted on a Windows file server (Storage Replica) running on flash storage (Pure). Our VMs are 2vCPU/4GB. We see about a ~3-5 second login impact to FlexEngine processing. I've found that ideally you want to run as much as you can asynchronously, and for shortcuts make sure you're referencing local ICO files on the template, not ones on a share (we just make a C:\Icons\ folder). Also eliminate as much from the user policy section of your GPO as you can, which should be pretty much everything except for the DEM config ADMX.

[sjesse]

Thanks for the reply, are you using windows 10 yet, if so I'm surprised your getting away with 4gb of memory. I've been doing testing and I've been seeing cpu spikes on logon that are lowered by using more memory.

I agree with the GPOs, I'm even using the NOAD mode in UEM because of our need to support multiple domains. We put everything in UEM and for the most part I've been happy, but our login times are a bit high, which is why I'm trying again to optimize each process for each of our images.

Good tip on the icons folder, I never thought of that and was accepting the blank shortcuts we've seen in some cases.

[valkyr0]

Yes, 4GB on Win10. Primary use case is just running an EHR, which is fairly lightweight. Some users could use 6GB.

As for login times, do you have verbose GPO processing on? If so, do you see what stage is taking the longest? "Applying UEM FlexEngine Policy" shouldn't be taking very long. Our login times with Win10 1709 are ~18 seconds w/ UEM. In environments with AppVolumes that's ~30 seconds. Mandatory profiles & ran OSOT (on the ntuser.man).

[DEMdev]

Hi sjesse,

Sorry that it took a while – a bit busy at the moment.

Looking at the two recent path-based imports in your log, I'd say both are relatively fast. One took 4421 ms, and the other 2272 ms. The largest difference between those two pretty much identical imports is in the import of profile archive 'Shared Settings.zip'. In the fast case, that takes 308 ms; in the slow one 1392 ms. Many small differences in other imports, but nothing sticking out.

This is maybe just the variability of an RDSH box?

[sjesse]

No worries, bet you are making the DEM product even better .  Thanks for catching that I didn't even notice, I'm hoping once I build updated fileservers maybe that would help. I just setup our new equipment but I only have this one 2012 r2 server on these massive 40 core blades. Theres nothing else so I can't think of a reason why it would be intermittent, other then the infrastructure servers are in another enclosure on 8 year old hardware. I'm hoping once I get the new file servers up and point uem to them, maybe I'd see improvement because they traffic wouldn't be that far away, and on more up to date hardware. Figure now is the time to focus on improvment.

[DEMdev]

Hi sjesse,

Yep, busy with 9.10

In 9.9 we added configuration options to enable some diagnostics. Maybe useful for trying to figure out what's different from one session to the next? It's not like these are high fidelity performance measurement features, but maybe they can help you spot some outliers.