[dpdk-dev] [PATCH v2] app/testpmd: add mempool bulk get for txonly mode

[Pavan Nikhilesh Bhagavatula]

> -----Original Message-----
> From: Iremonger, Bernard <bernard.iremonger at intel.com>
> Sent: Tuesday, March 26, 2019 5:20 PM
> To: Thomas Monjalon <thomas at monjalon.net>; Pavan Nikhilesh
> Bhagavatula <pbhagavatula at marvell.com>
> Cc: dev at dpdk.org; Jerin Jacob Kollanukkaran <jerinj at marvell.com>;
> arybchenko at solarflare.com; Yigit, Ferruh <ferruh.yigit at intel.com>
> Subject: [EXT] RE: [dpdk-dev] [PATCH v2] app/testpmd: add mempool bulk
> get for txonly mode
> 
> External Email
> 
> ----------------------------------------------------------------------
> Hi Pavan,
> 
> <snip>
> 
> > Subject: Re: [dpdk-dev] [PATCH v2] app/testpmd: add mempool bulk get
> > for txonly mode
> >
> > 01/03/2019 14:47, Pavan Nikhilesh Bhagavatula:
> > > From: Pavan Nikhilesh <pbhagavatula at marvell.com>
> > >
> > > Use mempool bulk get ops to alloc burst of packets and process them.
> > > If bulk get fails fallback to rte_mbuf_raw_alloc.
> > >
> > > Suggested-by: Andrew Rybchenko <arybchenko at solarflare.com>
> > > Signed-off-by: Pavan Nikhilesh <pbhagavatula at marvell.com>
> > > ---
> > >
> > >  v2 Changes:
> > >  - Use bulk ops for fetching segments. (Andrew Rybchenko)
> > >  - Fallback to rte_mbuf_raw_alloc if bulk get fails. (Andrew
> > > Rybchenko)
> > >  - Fix mbufs not being freed when there is no more mbufs available
> > > for segments. (Andrew Rybchenko)
> > >
> > >  app/test-pmd/txonly.c | 159
> > > ++++++++++++++++++++++++------------------
> > >  1 file changed, 93 insertions(+), 66 deletions(-)
> >
> > This is changing a lot of lines so it is difficult to know what is changed
> exactly.
> > Please split it with a refactoring without any real change, and
> > introduce the real change later.

Ok will split in the next version.

> > Then we'll be able to examine it and check the performance.
> >
> > We need to have more tests with more hardware in order to better
> > understand the performance improvement.
> > For info, a degradation is seen in Mellanox lab.

The only real change was that we introduced mempool_bulk get to avoid multiple calls to mempool layer. 
I don't see how that would degrade the performance.

> >
> >
> +1
> Not easy to review.
> Btw, unnecessary change at lines  157 and 158 in txonly.c

Will remove the assignments.

> 
> Regards,
> 
> Bernard