It would be nice for Nerdio to be able to have the ability to right-size VM's based on metrics (CPU or RAM usage)
Not sure if this should be an automatic thing or if it should alert you and require consent before resizing the VM.
This could work on both host pools and servers.
Not just by halving, or doubling, but also by family. Maybe a D-series would work better than an E-series because you didn't need as much RAM as you thought you did, and you are leaving money on the table. Or you find that a B would work better than a D for your domain controllers.
This kind of feature aligns with Nerdio's primary practice of making AVD easy and affordable. It takes a lot of the guess work out of several builds as every environment is different.
Server Right-sizing (Planned)
6
Right Sizing is one of those things we get asked about pretty regularly in our Azure Business Reviews (ABRs).
Automating it (without consent) makes me a little nervous.
Maybe if we add an AI prompt:
“We've noticed <ServerName> is under-utilized. We recommend changing to an D2as_v6 for the best cost to performance ratio. Would you like to have that automatically changed during your scheduled maintenance window (Tuesday at 2AM)?”
That way, you can get the automation with less risk than automation or manual notification and manual adjustment.
Thoughts?
For sure!
I think if we start small, like AI suggestions, and allow our fellow NMM users chime in with requests that piggyback off this feature, we might be able to fine tune it to make it work for the general masses.
As an optional advisory feature, it seems like a solid idea.
One thing that might be key to flag with that recommendation is how the recommended change would impact scale out numbers.
We could easily run on a single D2ads_v5 off-hours instead of a D8ads_v5, but that would also change our scale out VMs from D8ads to D2ads which would make the pool grow from 10 peak hosts to 40+.
For some customers (we'd be in this category) it would also be helpful to be able to set the horizon that tool calculates over. We have a cyclical user count with notable peaks at end of month so a 7 or 14 day calculation would likely recommend an unnecessary downsize with associated increase to the number of scaled out hosts in the pool.
Thanks, Brian. 🙂
That kind of logic may be difficult to implement. Especially with Always On/Reserved Instance hosts.
I think that Jason was talking about resizing Server VMs and not necessarily session hosts (although I could be wrong), but his other idea (Max Pool Limit Size – Nerdio Help Center) may do what you're talking about, so you don't run into expensive overprovisioning.
One item that we would find helpful here to optimize costs would be having the ability to run a differently sized host for the always-on from the scale out VMs. I'd love to have the ability to set:
- Always on VM: D2ads
- Scale out VMs: D8ads
Please sign in to leave a comment.
Comments (5 comments)