[ovs-dev] [PATCH 2/8] doc: Add "PMD" topic document

Stephen Finucane stephen at that.guru
Mon Apr 16 13:18:40 UTC 2018


On Mon, 2018-04-09 at 15:16 +0000, Stokes, Ian wrote:
> > This continues the breakup of the huge DPDK "howto" into smaller
> > components. There are a couple of related changes included, such as using
> > "Rx queue" instead of "rxq" and noting how Tx queues cannot be configured.
> > 
> > We enable the TODO directive, so we can actually start calling out some
> > TODOs.
> > 
> > Signed-off-by: Stephen Finucane <stephen at that.guru>
> > ---
> >  Documentation/conf.py                    |   2 +-
> >  Documentation/howto/dpdk.rst             |  86 -------------------
> >  Documentation/topics/dpdk/index.rst      |   1 +
> >  Documentation/topics/dpdk/phy.rst        |  10 +++
> >  Documentation/topics/dpdk/pmd.rst        | 139
> > +++++++++++++++++++++++++++++++
> >  Documentation/topics/dpdk/vhost-user.rst |  17 ++--
> >  6 files changed, 159 insertions(+), 96 deletions(-)  create mode 100644
> > Documentation/topics/dpdk/pmd.rst
> > 
> > diff --git a/Documentation/conf.py b/Documentation/conf.py index
> > 6ab144c5d..babda21de 100644
> > --- a/Documentation/conf.py
> > +++ b/Documentation/conf.py
> > @@ -32,7 +32,7 @@ needs_sphinx = '1.1'
> >  # Add any Sphinx extension module names here, as strings. They can be  #
> > extensions coming with Sphinx (named 'sphinx.ext.*') or your custom  #
> > ones.
> > -extensions = []
> > +extensions = ['sphinx.ext.todo']
> > 
> >  # Add any paths that contain templates here, relative to this directory.
> >  templates_path = ['_templates']
> > diff --git a/Documentation/howto/dpdk.rst b/Documentation/howto/dpdk.rst
> > index d717d2ebe..c2324118d 100644
> > --- a/Documentation/howto/dpdk.rst
> > +++ b/Documentation/howto/dpdk.rst
> > @@ -81,92 +81,6 @@ To stop ovs-vswitchd & delete bridge, run::
> >      $ ovs-appctl -t ovsdb-server exit
> >      $ ovs-vsctl del-br br0
> > 
> > -PMD Thread Statistics
> > ----------------------
> > -
> > -To show current stats::
> > -
> > -    $ ovs-appctl dpif-netdev/pmd-stats-show
> > -
> > -To clear previous stats::
> > -
> > -    $ ovs-appctl dpif-netdev/pmd-stats-clear
> > -
> > -Port/RXQ Assigment to PMD Threads
> > ----------------------------------
> > -
> > -To show port/rxq assignment::
> > -
> > -    $ ovs-appctl dpif-netdev/pmd-rxq-show
> > -
> > -To change default rxq assignment to pmd threads, rxqs may be manually
> > pinned to -desired cores using::
> > -
> > -    $ ovs-vsctl set Interface <iface> \
> > -        other_config:pmd-rxq-affinity=<rxq-affinity-list>
> > -
> > -where:
> > -
> > -- ``<rxq-affinity-list>`` is a CSV list of ``<queue-id>:<core-id>``
> > values
> > -
> > -For example::
> > -
> > -    $ ovs-vsctl set interface dpdk-p0 options:n_rxq=4 \
> > -        other_config:pmd-rxq-affinity="0:3,1:7,3:8"
> > -
> > -This will ensure:
> > -
> > -- Queue #0 pinned to core 3
> > -- Queue #1 pinned to core 7
> > -- Queue #2 not pinned
> > -- Queue #3 pinned to core 8
> > -
> > -After that PMD threads on cores where RX queues was pinned will become -
> > ``isolated``. This means that this thread will poll only pinned RX queues.
> > -
> > -.. warning::
> > -  If there are no ``non-isolated`` PMD threads, ``non-pinned`` RX queues
> > will
> > -  not be polled. Also, if provided ``core_id`` is not available (ex. this
> > -  ``core_id`` not in ``pmd-cpu-mask``), RX queue will not be polled by
> > any PMD
> > -  thread.
> > -
> > -If pmd-rxq-affinity is not set for rxqs, they will be assigned to pmds
> > (cores) -automatically. The processing cycles that have been stored for
> > each rxq -will be used where known to assign rxqs to pmd based on a round
> > robin of the -sorted rxqs.
> > -
> > -For example, in the case where here there are 5 rxqs and 3 cores (e.g.
> > 3,7,8) -available, and the measured usage of core cycles per rxq over the
> > last -interval is seen to be:
> > -
> > -- Queue #0: 30%
> > -- Queue #1: 80%
> > -- Queue #3: 60%
> > -- Queue #4: 70%
> > -- Queue #5: 10%
> > -
> > -The rxqs will be assigned to cores 3,7,8 in the following order:
> > -
> > -Core 3: Q1 (80%) |
> > -Core 7: Q4 (70%) | Q5 (10%)
> > -core 8: Q3 (60%) | Q0 (30%)
> > -
> > -To see the current measured usage history of pmd core cycles for each
> > rxq::
> > -
> > -    $ ovs-appctl dpif-netdev/pmd-rxq-show
> > -
> > -.. note::
> > -
> > -  A history of one minute is recorded and shown for each rxq to allow for
> > -  traffic pattern spikes. An rxq's pmd core cycles usage changes due to
> > traffic
> > -  pattern or reconfig changes will take one minute before they are fully
> > -  reflected in the stats.
> > -
> > -Rxq to pmds assignment takes place whenever there are configuration
> > changes -or can be triggered by using::
> > -
> > -    $ ovs-appctl dpif-netdev/pmd-rxq-rebalance
> > -
> >  QoS
> >  ---
> > 
> > diff --git a/Documentation/topics/dpdk/index.rst
> > b/Documentation/topics/dpdk/index.rst
> > index 5f836a6e9..dfde88377 100644
> > --- a/Documentation/topics/dpdk/index.rst
> > +++ b/Documentation/topics/dpdk/index.rst
> > @@ -31,3 +31,4 @@ The DPDK Datapath
> >     phy
> >     vhost-user
> >     ring
> > +   pmd
> > diff --git a/Documentation/topics/dpdk/phy.rst
> > b/Documentation/topics/dpdk/phy.rst
> > index 1c18e4e3d..222fa3e9f 100644
> > --- a/Documentation/topics/dpdk/phy.rst
> > +++ b/Documentation/topics/dpdk/phy.rst
> > @@ -109,3 +109,13 @@ tool::
> >  For more information, refer to the `DPDK documentation <dpdk-drivers>`__.
> > 
> >  .. _dpdk-drivers: http://dpdk.org/doc/guides/linux_gsg/linux_drivers.html
> > +
> > +Multiqueue
> > +----------
> > +
> > +Poll Mode Driver (PMD) threads are the threads that do the heavy
> > +lifting for the DPDK datapath. Correct configuration of PMD threads and
> > +the Rx queues they utilize is a requirement in order to deliver the
> > +high-performance possible with the DPDK datapath. It is possible to
> > +configure multiple Rx queues for ``dpdk`` ports, thus ensuring this is
> > +not a bottleneck for performance. For information on configuring PMD
> > threads, refer to :doc:`pmd`.
> > diff --git a/Documentation/topics/dpdk/pmd.rst
> > b/Documentation/topics/dpdk/pmd.rst
> > new file mode 100644
> > index 000000000..e15e8cc3b
> > --- /dev/null
> > +++ b/Documentation/topics/dpdk/pmd.rst
> > @@ -0,0 +1,139 @@
> > +..
> > +      Licensed under the Apache License, Version 2.0 (the "License"); you
> > may
> > +      not use this file except in compliance with the License. You may
> > obtain
> > +      a copy of the License at
> > +
> > +          http://www.apache.org/licenses/LICENSE-2.0
> > +
> > +      Unless required by applicable law or agreed to in writing, software
> > +      distributed under the License is distributed on an "AS IS" BASIS,
> > WITHOUT
> > +      WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
> > See the
> > +      License for the specific language governing permissions and
> > limitations
> > +      under the License.
> > +
> > +      Convention for heading levels in Open vSwitch documentation:
> > +
> > +      =======  Heading 0 (reserved for the title in a document)
> > +      -------  Heading 1
> > +      ~~~~~~~  Heading 2
> > +      +++++++  Heading 3
> > +      '''''''  Heading 4
> > +
> > +      Avoid deeper levels because they do not render well.
> > +
> > +===========
> > +PMD Threads
> > +===========
> > +
> > +Poll Mode Driver (PMD) threads are the threads that do the heavy
> > +lifting for the DPDK datapath and perform tasks such as continuous
> > +polling of input ports for packets, classifying packets once received,
> > +and executing actions on the packets once they are classified.
> > +
> > +PMD threads utilize Receive (Rx) and Transmit (Tx) queues, commonly
> > +known as *rxq*\s and *txq*\s. While Tx queue configuration happens
> > +automatically, Rx queues can be configured by the user. This can happen
> > in one of two ways:
> 
> Just on above, could be a to-do but it's a good opportunity to add a
> note on the "automatic" behavior of tx queues, number created and how
> it relates to the number of PMDs etc. Could be a separate section in
> the PMD doc.

Yeah, if it's OK with you I'll add this as a TODO and then work with
you to write this additional section.

> > +
> > +- For physical interfaces, configuration is done using the
> > +  :program:`ovs-appctl` utility.
> > +
> > +- For virtual interfaces, configuration is done using the
> > +:program:`ovs-appctl`
> > +  utility, but this configuration must be reflected in the guest
> > +configuration
> > +  (e.g. QEMU command line arguments).
> > +
> > +The :program:`ovs-appctl` utility also provides a number of commands
> > +for querying PMD threads and their respective queues. This, and all of
> > +the above, is discussed here.
> > +
> > +PMD Thread Statistics
> > +---------------------
> > +
> > +To show current stats::
> > +
> > +    $ ovs-appctl dpif-netdev/pmd-stats-show
> > +
> > +To clear previous stats::
> > +
> > +    $ ovs-appctl dpif-netdev/pmd-stats-clear
> > +
> > +Port/Rx Queue Assigment to PMD Threads
> > +--------------------------------------
> > +
> > +.. todo::
> > +
> > +   This needs a more detailed overview of *why* this should be done,
> > along with
> > +   the impact on things like NUMA affinity.
> > +
> > +To show port/RX queue assignment::
> > +
> > +    $ ovs-appctl dpif-netdev/pmd-rxq-show
> > +
> > +Rx queues may be manually pinned to cores. This will change the default
> > +Rx queue assignment to PMD threads::
> > +
> > +    $ ovs-vsctl set Interface <iface> \
> > +        other_config:pmd-rxq-affinity=<rxq-affinity-list>
> > +
> > +where:
> > +
> > +- ``<rxq-affinity-list>`` is a CSV list of ``<queue-id>:<core-id>``
> > +values
> > +
> > +For example::
> > +
> > +    $ ovs-vsctl set interface dpdk-p0 options:n_rxq=4 \
> > +        other_config:pmd-rxq-affinity="0:3,1:7,3:8"
> > +
> > +This will ensure there are *4* Rx queues and that these queues are
> > +configured like so:
> > +
> > +- Queue #0 pinned to core 3
> > +- Queue #1 pinned to core 7
> > +- Queue #2 not pinned
> > +- Queue #3 pinned to core 8
> > +
> > +PMD threads on cores where Rx queues are *pinned* will become
> > +*isolated*. This means that this thread will only poll the *pinned* Rx
> > queues.
> > +
> > +.. warning::
> > +
> > +  If there are no *non-isolated* PMD threads, *non-pinned* RX queues
> > + will not  be polled. Also, if the provided ``<core-id>`` is not
> > + available (e.g. the  ``<core-id>`` is not in ``pmd-cpu-mask``), the RX
> > + queue will not be polled by  any PMD thread.
> > +
> > +If ``pmd-rxq-affinity`` is not set for Rx queues, they will be assigned
> > +to PMDs
> > +(cores) automatically. Where known, the processing cycles that have
> > +been stored for each Rx queue will be used to assign Rx queue to PMDs
> > +based on a round robin of the sorted Rx queues. For example, take the
> > +following example, where there are five Rx queues and three cores - 3,
> > +7, and 8 - available and the measured usage of core cycles per Rx queue
> > +over the last interval is seen to
> > +be:
> > +
> > +- Queue #0: 30%
> > +- Queue #1: 80%
> > +- Queue #3: 60%
> > +- Queue #4: 70%
> > +- Queue #5: 10%
> > +
> > +The Rx queues will be assigned to the cores in the following order:
> > +
> > +Core 3: Q1 (80%) |
> > +Core 7: Q4 (70%) | Q5 (10%)
> > +core 8: Q3 (60%) | Q0 (30%)
> > +
> 
> This functionality was introduced in OVS 2.8. Do we need to warn the
> user with a versionchanged:: 2.8.0 and that it's unavailable prior to
> this? The behavior in that case was round robin without taking
> processing cycles into consideration. There would also be no history
> tracking for the stats and no pmd rebalance command.

Yes, I'll add this.

> > +To see the current measured usage history of PMD core cycles for each
> > +Rx
> > +queue::
> > +
> > +    $ ovs-appctl dpif-netdev/pmd-rxq-show
> > +
> > +.. note::
> > +
> > +  A history of one minute is recorded and shown for each Rx queue to
> > + allow for  traffic pattern spikes. Any changes in the Rx queue's PMD
> > + core cycles usage,  due to traffic pattern or reconfig changes, will
> > + take one minute to be fully  reflected in the stats.
> > +
> > +Rx queue to PMD assignment takes place whenever there are configuration
> > +changes or can be triggered by using::
> > +
> > +    $ ovs-appctl dpif-netdev/pmd-rxq-rebalance
> 
> We should probably flag to users considerations for PMD and multi queue specific to phy and vhost ports.
> 
> Perhaps a link to the specific documents below along with the heads up:
> 
> Documentation/topics/dpdk/vhost-user.rst
> Documentation/topics/dpdk/phy.rst

Yup, good call. Done.

Stephen

> Ian
> 
> > diff --git a/Documentation/topics/dpdk/vhost-user.rst
> > b/Documentation/topics/dpdk/vhost-user.rst
> > index 95517a676..d84d99246 100644
> > --- a/Documentation/topics/dpdk/vhost-user.rst
> > +++ b/Documentation/topics/dpdk/vhost-user.rst
> > @@ -127,11 +127,10 @@ an additional set of parameters::
> >      -netdev type=vhost-user,id=mynet2,chardev=char2,vhostforce
> >      -device virtio-net-pci,mac=00:00:00:00:00:02,netdev=mynet2
> > 
> > -In addition,       QEMU must allocate the VM's memory on hugetlbfs.
> > vhost-user
> > -ports access a virtio-net device's virtual rings and packet buffers
> > mapping the -VM's physical memory on hugetlbfs. To enable vhost-user ports
> > to map the VM's -memory into their process address space, pass the
> > following parameters to
> > -QEMU::
> > +In addition, QEMU must allocate the VM's memory on hugetlbfs.
> > +vhost-user ports access a virtio-net device's virtual rings and packet
> > +buffers mapping the VM's physical memory on hugetlbfs. To enable
> > +vhost-user ports to map the VM's memory into their process address space,
> > pass the following parameters to QEMU::
> > 
> >      -object memory-backend-file,id=mem,size=4096M,mem-
> > path=/dev/hugepages,share=on
> >      -numa node,memdev=mem -mem-prealloc @@ -151,18 +150,18 @@ where:
> >    The number of vectors, which is ``$q`` * 2 + 2
> > 
> >  The vhost-user interface will be automatically reconfigured with required
> > -number of rx and tx queues after connection of virtio device.  Manual
> > +number of Rx and Tx queues after connection of virtio device.  Manual
> >  configuration of ``n_rxq`` is not supported because OVS will work
> > properly only  if ``n_rxq`` will match number of queues configured in
> > QEMU.
> > 
> > -A least 2 PMDs should be configured for the vswitch when using
> > multiqueue.
> > +A least two PMDs should be configured for the vswitch when using
> > multiqueue.
> >  Using a single PMD will cause traffic to be enqueued to the same vhost
> > queue  rather than being distributed among different vhost queues for a
> > vhost-user  interface.
> > 
> >  If traffic destined for a VM configured with multiqueue arrives to the
> > vswitch -via a physical DPDK port, then the number of rxqs should also be
> > set to at -least 2 for that physical DPDK port. This is required to
> > increase the
> > +via a physical DPDK port, then the number of Rx queues should also be
> > +set to at least two for that physical DPDK port. This is required to
> > +increase the
> >  probability that a different PMD will handle the multiqueue transmission
> > to the  guest using a different vhost queue.
> > 
> > --
> > 2.14.3
> > 
> > _______________________________________________
> > dev mailing list
> > dev at openvswitch.org
> > https://mail.openvswitch.org/mailman/listinfo/ovs-dev



More information about the dev mailing list