[ovs-dev] [v4 03/12] dpif-netdev: Add study function to select the best mfex function

Van Haaren, Harry harry.van.haaren at intel.com
Wed Jun 30 13:34:12 UTC 2021

> -----Original Message-----
> From: Eelco Chaudron <echaudro at redhat.com>
> Sent: Wednesday, June 30, 2021 11:07 AM
> To: Van Haaren, Harry <harry.van.haaren at intel.com>
> Cc: Flavio Leitner <fbl at sysclose.org>; Amber, Kumar <kumar.amber at intel.com>;
> dev at openvswitch.org; i.maximets at ovn.org
> Subject: Re: [ovs-dev] [v4 03/12] dpif-netdev: Add study function to select the best
> mfex function
> On 30 Jun 2021, at 11:43, Van Haaren, Harry wrote:


> > $ ovs-appctl  dpif-netdev/miniflow-parser-set avx512_ipv4_udp
> >
> > There is an assumption here that all datapath threads handle
> > the same outer traffic type. If that's not the case, we cannot manually
> > set different MFEX impls to different pmd threads today, as your lab
> > to production requirement requests above.
> >
> > If we add an optional PMD thread id parameter, we can support this:
> > $ ovs-appctl  dpif-netdev/miniflow-parser-set avx512_ipv4_udp
> <packet_count_to_study> <pmd thread idx>
> I think if we allow study to set it per PMD thread, we should support the pmd thread
> for manual configuration.
> We also might need to re-think the command to make sure packet_count_to_study
> is only needed for the study command.
> So the help text might become something like:
> dpif-netdev/miniflow-parser-set {miniflow_implementation_name | study [pkt_cnt]}
> [dp] [pmd_core]

Amber has designed & implemented a proposal, with documentation on each. Request to
review the next version of the patchset when available, to ensure it meets requirements.

More information about the dev mailing list