[dpdk-dev] [PATCH v5] app/testpmd: add option ring-bind-lcpu to bind Q with CPU

[Simon Guo]

Hi Konstantin,
On Thu, Jan 18, 2018 at 12:14:05PM +0000, Ananyev, Konstantin wrote:
> Hi Simon,
> 
> > 
> > Hi, Konstantin,
> > On Tue, Jan 16, 2018 at 12:38:35PM +0000, Ananyev, Konstantin wrote:
> > >
> > >
> > > > -----Original Message-----
> > > > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of wei.guo.simon at gmail.com
> > > > Sent: Saturday, January 13, 2018 2:35 AM
> > > > To: Lu, Wenzhuo <wenzhuo.lu at intel.com>
> > > > Cc: dev at dpdk.org; Thomas Monjalon <thomas at monjalon.net>; Simon Guo <wei.guo.simon at gmail.com>
> > > > Subject: [dpdk-dev] [PATCH v5] app/testpmd: add option ring-bind-lcpu to bind Q with CPU
> > > >
> > > > From: Simon Guo <wei.guo.simon at gmail.com>
> > > >
> > > > Currently the rx/tx queue is allocated from the buffer pool on socket of:
> > > > - port's socket if --port-numa-config specified
> > > > - or ring-numa-config setting per port
> > > >
> > > > All the above will "bind" queue to single socket per port configuration.
> > > > But it can actually archieve better performance if one port's queue can
> > > > be spread across multiple NUMA nodes, and the rx/tx queue is allocated
> > > > per lcpu socket.
> > > >
> > > > This patch adds a new option "--ring-bind-lcpu"(no parameter).  With
> > > > this, testpmd can utilize the PCI-e bus bandwidth on another NUMA
> > > > nodes.
> > > >
> > > > When --port-numa-config or --ring-numa-config option is specified, this
> > > > --ring-bind-lcpu option will be suppressed.
> > >
> > > Instead of introducing one more option - wouldn't it be better to
> > > allow user manually to define flows and assign them to particular lcores?
> > > Then the user will be able to create any FWD configuration he/she likes.
> > > Something like:
> > > lcore X add flow rxq N,Y txq M,Z
> > >
> > > Which would mean - on lcore X recv packets from port=N, rx_queue=Y,
> > > and send them through port=M,tx_queue=Z.
> > Thanks for the comment.
> > Will it be a too compliated solution for user since it will need to define
> > specifically for each lcore? We might have hundreds of lcores in current
> > modern platforms.
> 
> Why for all lcores?
> Only for ones that will do packet forwarding.
> Also if configuration becomes too complex(/big) to be done manually
> user can write a script that will generate set of testpmd commands
> to achieve desired layout.

It might not be an issue for skillful users, but it will be difficult
for others.  --ring-bind-lcpu will help to simply this for them.

Thanks,
- Simon