[ovs-dev] [PATCH RFC v3] dpif-netdev: Allow different numbers of rx queues for different ports.

[Flavio Leitner]

On Thu, 21 Jan 2016 17:15:18 +0300
Ilya Maximets <i.maximets at samsung.com> wrote:

> Currently, all of the PMD netdevs can only have the same number of
> rx queues, which is specified in other_config:n-dpdk-rxqs.
> 
> Fix that by introducing of new option for PMD interfaces: 'n_rxq',
> which specifies the maximum number of rx queues to be created for this
> interface.
> 
> Example:
> 	ovs-vsctl set Interface dpdk0 options:n_rxq=8
> 
> Old 'other_config:n-dpdk-rxqs' deleted.
> 
> Signed-off-by: Ilya Maximets <i.maximets at samsung.com>
> Acked-by: Ben Pfaff <blp at ovn.org>
> ---
> Version 3:
> 	* Added info to NEWS.
> 	* Fixed INSTALL.DPDK.md
> 	* Fixed checking 'netdev_n_rxq() != netdev_requested_n_rxq()'
> 
> Version 2:
> 	* Changed wrong comment in struct netdev as suggested by Ben
> Pfaff.
> 
>  INSTALL.DPDK.md            |  4 ++--
>  NEWS                       |  4 ++++
>  lib/dpif-netdev.c          | 51
> +++++++++++++++++++++++++++-------------------
> lib/dpif-provider.h        |  8 +++----- lib/dpif.c
> |  5 ++--- lib/dpif.h                 |  3 +--
>  lib/netdev-dpdk.c          | 26 ++++++++++++++++++-----
>  lib/netdev-provider.h      |  6 +++++-
>  lib/netdev.c               |  7 +++++++
>  lib/netdev.h               |  1 +
>  ofproto/ofproto-dpif.c     |  2 +-
>  ofproto/ofproto-provider.h |  3 ---
>  ofproto/ofproto.c          |  7 -------
>  ofproto/ofproto.h          |  1 -
>  vswitchd/bridge.c          |  2 --
>  vswitchd/vswitch.xml       | 24 ++++++++++++++--------
>  16 files changed, 92 insertions(+), 62 deletions(-)

My concern is that this can create some interesting issues like
having a port with N queues for RX/TX, but only one operating with
many streams because the receiving NIC is single queue.

It also can create an unbalanced workload since you can have one
full device in one PMD plus queues of others devices.

Those aren't new problems, but having devices with different numbers
of queues can help those to happen more easily.

In any case, this seems to be the right direction to go.

The patch looks good and it works for me.

Acked-by: Flavio Leitner <fbl at sysclose.org>

Thanks,
-- 
fbl