Index | Thread | Search

From:
"Brian Conway" <bconway@rcesoftware.com>
Subject:
Re: vio(4): tso
To:
"Jan Klemkow" <j.klemkow@wemelug.de>
Cc:
tech@openbsd.org
Date:
Mon, 08 Apr 2024 07:52:00 -0500

Download raw body.

Thread
  • jan@openbsd.org:

    vio(4): tso

    • jan@openbsd.org:

      vio(4): tso

      • Brian Conway:

        vio(4): tso

      • Alexander Bluhm:

        vio(4): tso

On Fri, Apr 5, 2024, at 1:51 PM, Jan Klemkow wrote:
> On Sun, Mar 03, 2024 at 05:13:26PM +0100, jan@openbsd.org wrote:
>> On Thu, Feb 29, 2024 at 07:52:23AM -0600, Brian Conway wrote:
>> > Original thread, sorry for breaking threading: https://marc.info/?l=openbsd-tech&m=170523759201797
>> > 
>> > Greetings. Are you still looking for testing on this? I went to give
>> > it a spin today, but it no longer applies cleanly due to other recent
>> > vio(4) work.
>> 
>> I will update and resend the diff after the release of 7.5.  So, there
>> it more time for testing.  Before the 7.6 release.
>
> Here is an updated version of the vio(4) TSO diff.  It takes some
> advantage of the ether_extract_headers() improvements.  vlan(4) in
> combination with vio(4) will work, but does not use TSO for now.  Thus,
> no performance increase with vlan traffic.
>
> I tested it with IPv4 and IPv6 on a Linux KVM host.
>
> Tests are welcome.

I tested over the weekend and saw a minor but consistent improvement in send speed on my particular/arbitrary vmd host -> guest -> receive system path (725-730 Mbps improved to 760-770 Mbps).

Beyond that, no unusual instability or errors noticed. Hope that helps.

Brian Conway
Owner
RCE Software, LLC