[ovs-dev] [v6 06/11] dpif-netdev: Add packet count and core id paramters for study

[Stokes, Ian]

> > -----Original Message-----
> > From: Eelco Chaudron <echaudro at redhat.com>
> > Sent: Thursday, July 8, 2021 11:48 AM
> > To: Ferriter, Cian <cian.ferriter at intel.com>
> > Cc: ovs-dev at openvswitch.org; fbl at sysclose.org; i.maximets at ovn.org; Van
> > Haaren, Harry <harry.van.haaren at intel.com>; Amber, Kumar
> > <kumar.amber at intel.com>; Stokes, Ian <ian.stokes at intel.com>
> > Subject: Re: [v6 06/11] dpif-netdev: Add packet count and core id paramters
> for
> > study
> >
> > On 6 Jul 2021, at 15:11, Cian Ferriter wrote:
> >
> > > From: Kumar Amber <kumar.amber at intel.com>
> > >
> > > This commit introduces additional command line paramter
> > > for mfex study function. If user provides additional packet out
> > > it is used in study to compare minimum packets which must be processed
> > > else a default value is choosen.
> > > Also introduces a third paramter for choosing a particular pmd core.
> > >
> > > $ ovs-appctl dpif-netdev/miniflow-parser-set study 500 3
> > >
> > > Signed-off-by: Kumar Amber <kumar.amber at intel.com>
> >
> >
> > When doing some testing I ran into another corner case. If I start one PMD
> with
> > packet count 1000, and then another with packet count 10. Both will use 10,
> as
> > the packet count value, as is stored globally.
> 
> Valid corner case, good point, indeed the value is stored globally.
> 
> The packet-count for study was designed/added as a way to avoid studying
> too many (overhead) or too few (inaccuracy) packets. I'm not sure how useful
> it *actually is* in real-world.
> 
> Would documenting "the per packet count" is a global value, and parallel
> `study()`
> executions with differing packet counts will use the most recent count value" be
> acceptable?

+1 to this, as a initial step I think this would be ok, it can change if required in the future based on feedback from deployment.
We've seen similar approaches to features such as RXQ load balancing in OVS using global values for sample time periods, I believe this is a similar case.

Ian
> 
> In any given scenario, I expect the "pkt_cnt" value to be the same for all PMD
> threads as variance in traffic is likely to be constant for all PMD-threads
> (particularly
> since auto-load-balancing of rxq/pmds is being improved). A single global value
> being
> used covers all my known use-cases.
> 
> Thoughts?
> 
> > > ---
> > < SNIP>
>