On 11/12/17 17:31, Ilya Maximets wrote:
Added information about output packet batching and a way to
configure 'tx-flush-interval'.

Signed-off-by: Ilya Maximets <[email protected]>
---

This patch is made on top of v7 of Output Batching patch-set:

        https://mail.openvswitch.org/pipermail/ovs-dev/2017-December/341936.html

This patch sent as RFC separately from the rest of the series because
it's new and I don't want this to block accepting of the main series.
It'll be much easier to fix/rebase just this patch without re-sending
of the whole patch-set.

  Documentation/intro/install/dpdk.rst | 24 ++++++++++++++++++++++++
  1 file changed, 24 insertions(+)

diff --git a/Documentation/intro/install/dpdk.rst 
b/Documentation/intro/install/dpdk.rst
index 3fecb5c..b0bb76e 100644
--- a/Documentation/intro/install/dpdk.rst
+++ b/Documentation/intro/install/dpdk.rst
@@ -568,6 +568,30 @@ not needed i.e. jumbo frames are not needed, it can be 
forced off by adding
  chains of descriptors it will make more individual virtio descriptors 
available
  for rx to the guest using dpdkvhost ports and this can improve performance.
+Output Packet Batching
+~~~~~~~~~~~~~~~~~~~~~~
+
See some _XYZ_ remarks below, but I guess a native speaker might be better at this...
+To get advantages of _THE_ batched send functions OVS collects packets in
+intermediate queues before sending. This allows _USING_ _A_ single send for
+packets matched by different flows but having _THE_ same output action. 
Furthermore,
+OVS is able to collect packets for some reasonable amount of time before
+_BATCH_ sending _THEM_ _WHICH MIGHT HELP WHEN_ input batches are small.
+
+``tx-flush-interval`` config could be used to specify the time in microseconds
+that a packet can wait in an output queue for sending (default is ``0``)::
+
+    $ ovs-vsctl set Open_vSwitch . other_config:tx-flush-interval=50
+
+Lower values _DECREASE_ latency while higher values may be useful to achieve
+higher performance. For example, increasing of ``tx-flush-interval`` can be
+used to decrease _THE_ number of interrupts for interrupt based guest drivers.
+This may significantly affect the performance. Zero value means immediate
+send at the end of processing _DEL_of_ a single input batch.
+
+Average number of packets per output batch could be checked in PMD stats::
+
+    $ ovs-appctl dpif-netdev/pmd-stats-show
+
  Limitations
  ------------


_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to