VM packet loss when running multiple VMs simultaneously

[lynaghk]

Sometimes running VMs start dropping packets. The most obvious symptom is that codex/claude get into reconnect loops or just hang "working" without output for 20+ minutes.

I've confirmed packet loss by running ping 8.8.8.8 both in a VM and on the host; packets are only dropped from the former.

According to this issue there are sometimes problems related to TCP offloading.
I disabled that via:

ethtool -K enp0s1 tx off rx off tso off gso off gro off

but still had packet loss.

This issue is a nice overview of Apple Virtualization NAT problems and why bridged networking might be preferred.

Unfortunately bridged networking requires a restricted entitlement from Apple that I'm not going to get.

I've dug up some other projects that seem like they might be able to help, but I don't understand enough yet about networking to see what's up:

https://github.com/nirs/vmnet-helper
https://github.com/cirruslabs/softnet
https://github.com/edigaryev/vmnet
https://github.com/nagypeterjob/sock-vmnet
https://github.com/jdork0/vzcli

[lynaghk]

Integrated vmnet-helper in 7ac3ee2.

[lynaghk]

I've run into this problem again on both my Macbook Air M1 and Mini M4 (running MacOS 14 and 15, respectively) even when using the new vmnet-helper for NAT networking.