vRSS support on FreeBSD

[Wei Hu]

The table is sent through an out of band message with a specific head to the synthetic NIC. The driver on the guest side understands it and treats it as a update of send table.

Do you mean the host OS put the tx queue selection directly on the receive packet descriptor? It is a very good idea. The driver doesn't need to calculate the hash -- it already has the tx queue number saved. The host side make the rebalance decision anyway, why just tell the guest directly?  For those packets that don't have tx queen number it still can use the send table or default queue. 

Wei


-----Original Message-----
From: adrian.chadd at gmail.com [mailto:adrian.chadd at gmail.com] On Behalf Of Adrian Chadd
Sent: Wednesday, August 13, 2014 2:54 PM
To: Wei Hu
Cc: d at delphij.net; freebsd-net at freebsd.org
Subject: Re: vRSS support on FreeBSD

On 12 August 2014 23:27, Wei Hu <weh at microsoft.com> wrote:
> Unfortunately I am not aware of any official spec.

Damn!

Ok, so when they send the update for the redirect table, what's that look like? How do they send it?

> The driver for Linux guest OS does the similar thing for the tx path -- receive the send table update from host, calculate the hash value for each tx packet and use it to find the tx queue from the send table.
>
> I think the Windows guests do the same.

I wonder if there's any possibility for getting that added to the receive packet descriptor if the underlying hardware supports it - or well, if the OS has done it already.




-a