performance/limit of vm template references, how dynamically assign vgpu to vm

D

dfk1976

Active Member
Jan 5, 2020
21
3
43
48
I would like to know if this is possible with proxmox. Let's say we have a server with 4x RTX 5000 cards and also local U.3 SSD storage. Would it be possible to create one vm template and create 64 vms (based on this template) and can you also create these vms with dynamically assign a vgpu on creation? Or do you need to do this with a separate commandline or API call or is that also possible with the proxmox GUI? Are there any performance issues when using vm templates or are there any limits like for example maximum 8 vms per vm template? Any advices or experiences about this?
 
Last edited:
Not from gui but from shell / your own script. Your hw is the limit on how many created vm's from template you can run in cluster and no limit for a template as after they are independent after.
 
Ok, but if the hardware would run 64 vms, the question is if the usage of the vm template for as many vms will slow down the performance. So would it be better group/limit for example 8 vms per vm template? Or should you go with 64 vms using one base template?
 
dfk1976 said:
if the usage of the vm template for as many vms will slow down the performance
Click to expand...
MANY variable affect this. Ironically, if you're splitting your RTX5000 1:16 the gpu would NOT impact performance at all- each vm gets the same size shard. I dont really understand what you mean by template here.

capacity planning require you to measure you usecase for each vm in terms of:
cpu usage
ram usage
storage load
performance over time

map the load of 64 vms performance over time, and plan to have the overlapped peak of each type of resource as the MINIMUM you deploy.
 
This is an assumption to describe the problem. A template is a copy of an OS, for example Windows 10 as a golden image. If you now create many vms with this one template, differentiating images will be created. The question now is whether there is a performance loss if many vms only access one template or whether it is better to create several templates and, for example, only run 8 vms per template. Now do you see what I mean? Maybe someone here has experience as to which is better. Of course you can test it out when the time comes.
 
Last edited:
dfk1976 said:
If you now create many vms with this one template, differentiating images will be created.
Click to expand...
Oh you mean you want to use linked clones... that shouldnt have any impact on performance. Each vm will have its own disk for all subsequent writes, so performance will still be a consequence of your total write load and storage ability to perform under that load. this is the same whether you deploy each vm with its own copy or a linked master.
 
That's interesting, I wouldn't have thought that. So just one template (linked clone) is enough for many vms and it makes no difference in the SSD read write performance. Thank you for your info.
 
Last edited:
You must log in or register to reply here.
Top Bottom