To Reproduce Steps to reproduce the behavior:
Customer is unable to scale-out several different RHOSP 17.1 environments with IPv6 ctlplane networks and HP overcloud servers because of the same provisioning problem:

• provisioned node gets snponly.efi and boots it successfully
• downloads of boot.ipxe, kernel and initird are extremely slow and cause provisioning time out

We have collected tcpdumps for provisioning process and it looks like provisioned servers are processing data in extremely slow way: they ack small portions of received data, often ask to retransmit some of it.

We haven't ruled out networking issue completely yet because of inability to get consistent traffic dump on provisioned server's switchport; at the same time, same server is able to communicate with director properly when OS is booted from ISO. But we also need to take a look at this issue from bootloader/HW point of view. So I am reporting this issue and looking for your advice.

Expected behavior
Provisioned server is able to download kernel/initrd properly.

Bug impact
Scale-outs and node replacements are blocked in several RHOSP environments.

Known workaround
None