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 rxqs and txqs. While Tx queue configuration happens automatically, Rx queues can be configured by the user. This can happen in one of two ways:

  • For physical interfaces, configuration is done using the ovs-appctl utility.
  • For virtual interfaces, configuration is done using the ovs-appctl utility, but this configuration must be reflected in the guest configuration (e.g. QEMU command line arguments).

The 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

Correct configuration of PMD threads and the Rx queues they utilize is a requirement in order to achieve maximum performance. This is particularly true for enabling things like multiqueue for physical and vhost-user interfaces.

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.

The algorithm used to automatically assign Rxqs to PMDs can be set by:

$ ovs-vsctl set Open_vSwitch . other_config:pmd-rxq-assign=<assignment>

By default, cycles assignment is used where the Rxqs will be ordered by their measured processing cycles, and then be evenly assigned in descending order to PMDs based on an up/down walk of the PMDs. For 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%)

Alternatively, roundrobin assignment can be used, where the Rxqs are assigned to PMDs in a round-robined fashion. This algorithm was used by default prior to OVS 2.9. For example, given the following ports and queues:

  • Port #0 Queue #0 (P0Q0)
  • Port #0 Queue #1 (P0Q1)
  • Port #1 Queue #0 (P1Q0)
  • Port #1 Queue #1 (P1Q1)
  • Port #1 Queue #2 (P1Q2)

The Rx queues may be assigned to the cores in the following order:

Core 3: P0Q0 | P1Q1
Core 7: P0Q1 | P1Q2
Core 8: P1Q0 |

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

Changed in version 2.6.0: The pmd-rxq-show command was added in OVS 2.6.0.

Changed in version 2.9.0: Utilization-based allocation of Rx queues to PMDs and the pmd-rxq-rebalance command were added in OVS 2.9.0. Prior to this, allocation was round-robin and processing cycles were not taken into consideration.

In addition, the output of pmd-rxq-show was modified to include Rx queue utilization of the PMD as a percentage. Prior to this, tracking of stats was not available.