Traffic Shaping vs vmxnet3

[richardm]

I've seen some chatter on the 'net regarding traffic shaping not working well with the otherwise superior vmxnet3 interface.  Seems e1000e was/is the go-to interface for this scenario.  Where does this issue stand today?

[franco]

Which intel do you base this on?

Note that the shaper in OPNsense is what the limiter was in pfSense so what you heard about a shaper there does not apply here as we don't have ALTQ.


Cheers,
Franco

[richardm]

I remember seeing it when googling traffic shaper guidance and scenarios.  But I did read a lot of pfsense info and indeed this issue may be relevant to only pfsense (if at all).

What about TCP Segment Offload and other pNIC offloads at the VMware host level?  I've read that these can thwart various packet schedulers...

[franco]

The problem that we've seen is that these hardware accelerations don't get executed in the host leaving packets with wrong checksums to be discarded and causing horrible retransmission rates.

We've since disabled hardware acceleration for such reasons in the default install, but you're free to enable them bit by bit via Interfaces: Settings if you have a good test setup and chances are it works ok for VMware and Hyper-V.

Testing has the added benefit of knowing how much this really brings in terms of max. throughput. :)


Cheers,
Franco