CS Thesis using Incus cluster , would love some feedback

Would love to get anyone’s opinion on my CS Thesis.

Currently our cluster holds about 105 students that use it, not all at once though (not always at least).

I took some time to use Nvidia Scalable Functions (SF) for offloading network flows and RDMA for the storage replication. I feel like I tried to get data to prove my point that I could reach as close to line speed as possible with minimal to no resource load.

I tried to make the visualizations as easy to follow in the way I understood it. Im open to suggestions if you feel there could be more or less .

Anyways , any feedback helps.

RDMA_Accelerated_Container_Infrastructure.pdf (6.7 MB)

And this is the shortened down easier to follow and with all the same diagrams version to present with.

Presentation.pdf (6.0 MB)

I’d be curious to see what everyone has to think on whether or not I got my results because of the overkill of performance and it just happened to work in my favor? What would be my limit of users if I use offloaded interfaces as OVN uplinks before it starts slowing down?

@stgraber