Hi,
Note that I'm mostly speaking of theoretical limits and that there's currently no available Hardware which can top those, or even come close (besides custom-made super-computer stuff, obviously).
Currently: 8096 logical cores per host are the upper limit, 256 to 512 are rather the upper end of systems out in the wild nowadays.
~ 128 TiBs per node, as we currently use 4-level page tables (5 level would be possible).
Peta to Exabytes, depending on technology used. ZFS and Ceph should be both able to scale quite high here.
How many or how fast? Not sure if anybody run into a "how many" limit, this is normally limited by the count of PCIe lanes you can access on the motherboard. How fast? 100 GBps are working fine, those were used in a benchmark paper from 2020: (
https://forum.proxmox.com/threads/proxmox-ve-ceph-benchmark-2020-09-hyper-converged-with-nvme.76516/ ). There are 200 and 400 GBps, but they may need some tuning and fitting motherboard/CPU/... to work as advertised.
Do you mean nodes in a cluster? There are clusters with ~50 nodes out in the wild, but one may need good HW and network experience to go for over 16 and especially over 32 Nodes.
But remember, services like
Let's Encrypt (ACME) handle all their hundreds of millions of requests with a single server just fine, so with good powerful HW you can go a long way even when "just" having a 3 to 5 node cluster. Bonus: such setups are much less work to manage.
