Hi Ilya, I have applied your series cleanly on "[PATCH v9 0/2] Refactor PMD stats and cycle counting" and am testing this successfully on my node. So far the behavior is fully as expected and subjectively identical to the earlier versions I tested.
When running iperf UDP tests between VMs on different compute nodes I observe that the reported output packet batch size scales from 1 to 32 both with the offered load (packet rate) as well as the configured tx-flush-interval. The ping round-trip times are in line with the expectations from the configured tx-flush-interval. I will have a look at the individual patches and ack them. Regards, Jan > -----Original Message----- > From: Ilya Maximets [mailto:[email protected]] > Sent: Friday, 12 January, 2018 12:17 > To: [email protected] > Cc: Heetae Ahn <[email protected]>; Bhanuprakash Bodireddy > <[email protected]>; Antonio Fischetti > <[email protected]>; Eelco Chaudron <[email protected]>; Ciara > Loftus <[email protected]>; Kevin Traynor > <[email protected]>; Jan Scheurich <[email protected]>; Billy > O'Mahony <[email protected]>; Ian Stokes > <[email protected]>; Ilya Maximets <[email protected]> > Subject: [PATCH v10 0/5] Output packet batching (Time-based). > > This patch-set inspired by [1] from Bhanuprakash Bodireddy. > Implementation of [1] looks very complex and introduces many pitfalls [2] > for later code modifications like possible packet stucks. > > This version targeted to make simple and flexible output packet batching on > higher level without introducing and even simplifying netdev layer. > > [1] [PATCH v4 0/5] netdev-dpdk: Use intermediate queue during packet > transmission. > https://mail.openvswitch.org/pipermail/ovs-dev/2017-August/337019.html > > [2] For example: > https://mail.openvswitch.org/pipermail/ovs-dev/2017-August/337133.html > > The most of the series already in merge tree and not included in v9 and later > versions. > > Testing of 'PVP with OVS bonding on phy ports' scenario shows significant > performance improvement up to +25% (v9 has no performance difference with v6 > here): > > https://mail.openvswitch.org/pipermail/ovs-dev/2017-December/341700.html > > Other testing results for v6: > > https://mail.openvswitch.org/pipermail/ovs-dev/2017-December/341605.html > https://mail.openvswitch.org/pipermail/ovs-dev/2017-December/341628.html > > > Version 10: > * Rebased on top of '[PATCH v8 0/2] Refactor PMD stats and cycle > counting': > > https://mail.openvswitch.org/pipermail/ovs-dev/2018-January/343008.html > * Per-rxq cycles accounting changed to use new nested counters. > > Version 9: > * Made on top of current DPDK_MERGE tree (cc4891f). > * Patches 1-4,6 from v8 dropped as already included in DPDK_MERGE. > * "Time based ..." patch updated with per-rxq cycles counting from > the following incremental: > > https://mail.openvswitch.org/pipermail/ovs-dev/2017-December/342255.html > and splitted into 3 logically separate parts for easier review > purposes: > 1. Microsecond granularity for dpif-netdev. > 2. Cycles accounting changes. > 3. Time based batching itself. > There was no code changes in comparing with v8 + Incremental, only > re-split. > * Output batching mentioned in NEWS. > * Documentation patch included in this series. > > Version 8: > * Minor rebase on top of current master. > > Version 7: > * Rebased on current istokes/dpdk_merge (3eb8d4f) > * Dropped dp_netdev_pmd_thread structure refactoring patch. > Not needed since revert applied. > > Version 6: > * Rebased on current master: > - Added new patch to refactor dp_netdev_pmd_thread structure > according to following suggestion: > > https://mail.openvswitch.org/pipermail/ovs-dev/2017-November/341230.html > > NOTE: I still prefer reverting of the padding related patch. > Rebase done to not block acepting of this series. > Revert patch and discussion here: > > https://mail.openvswitch.org/pipermail/ovs-dev/2017-November/341153.html > > * Added comment about pmd_thread_ctx_time_update() usage. > > Version 5: > * pmd_thread_ctx_time_update() calls moved to different places to > call them only from dp_netdev_process_rxq_port() and main > polling functions: > pmd_thread_main, dpif_netdev_run and dpif_netdev_execute. > All other functions should use cached time from pmd->ctx.now. > It's guaranteed to be updated at least once per polling cycle. > * 'may_steal' patch returned to version from v3 because > 'may_steal' in qos is a completely different variable. This > patch only removes 'may_steal' from netdev API. > * 2 more usec functions added to timeval to have complete public API. > * Checking of 'output_cnt' turned to assertion. > > Version 4: > * Rebased on current master. > * Rebased on top of "Keep latest measured time for PMD thread." > (Jan Scheurich) > * Microsecond resolution related patches integrated. > * Time-based batching without RFC tag. > * 'output_time' renamed to 'flush_time'. (Jan Scheurich) > * 'flush_time' update moved to 'dp_netdev_pmd_flush_output_on_port'. > (Jan Scheurich) > * 'output-max-latency' renamed to 'tx-flush-interval'. > * Added patch for output batching statistics. > > Version 3: > > * Rebased on current master. > * Time based RFC: fixed assert on n_output_batches <= 0. > > Version 2: > > * Rebased on current master. > * Added time based batching RFC patch. > * Fixed mixing packets with different sources in same batch. > > > Ilya Maximets (5): > dpif-netdev: Use microsecond granularity. > dpif-netdev: Count cycles on per-rxq basis. > dpif-netdev: Time based output batching. > docs: Describe output packet batching in DPDK guide. > NEWS: Mark output packet batching support. > > Documentation/intro/install/dpdk.rst | 58 +++++++++++++ > NEWS | 2 + > lib/dpif-netdev.c | 164 > +++++++++++++++++++++++++++-------- > vswitchd/vswitch.xml | 16 ++++ > 4 files changed, 204 insertions(+), 36 deletions(-) > > -- > 2.7.4 _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
