[ovs-dev] OVS-DPDK bi-weekly meeting minutes

Stokes, Ian ian.stokes at intel.com
Fri Jul 14 11:31:59 UTC 2017


Meeting minutes. Feel free to correct/add.

14th July 2017

ATTENDEES
Ian Stokes, Antonio Fischetti, Billy O'Mahony, Eelco Chaudron, Flavio Leitner, Irene Liew, Jan Scheurich, Mark Kavanagh, Michelle Devlin, Ren Wang, Robin Giller, Rony Effron, Sugesh Chandran, Yipeng Wang, Frikkie Scholtz, Michael Lilja, Simon Horman, . People may have later in the meeting so I may have missed some people.


=======
GENERAL
=======

- OVS 2.8 Key Dates
-- When is branch date and release date?
-- No definite dates for this. General feeling is branch freeze is imminent however.
-- Once branch freeze is in place, no new features will be considered for 2.8
-- Existing patches that are under review will still be considered for the 2.8 release.

- OVS maintainer update
-- No one present able to provide update as regards this.

- Community etiquette and patch review process.
-- Issue raised regarding the patch review etiquette on the mailing list.
--- Recently examples of patches being blocked due to lack of feedback from reviewers who have previously flagged concerns with a patch.
--- General feeling is if concerns have been raised then feedback should be timely if the issues are addressed or under discussion to avoid blocking the patch.
--- If alternative approaches are possible people would like these to be discussed openly on the mailing list with whoever has been involved in the review up to that point.
--- Developing and delivering patches that achieve the same functionality but in a different manner without discussing the approach to an existing review seems counter intuitive to the community spirit.
-- Other open source groups have rules regarding reviews and process, is this something to be considered for OVS?
-- Require input from maintainers with regards to this issue.

====================
PERFORMANCE/FEATURES
====================
- Cuckoo distributor
-- Yipeng presented slides describing the behavior of the cuckoo distributor for prioritizing subtables in the classifier.
-- 2 questions raised with regards to this feature:
--- 1: How should the feature be disabled, it should be when the average look up of the sub table is 1.
--- 2: Question raised regarding how this work will help when the hash for a packet does not change when it is sent for recirculation?
---  When a packet is sent for recirculation the hash will change as the recirculation ID will be incremented and this is used in generating the new hash.
-- Antonio Fischetti has submitted RFC patch to skip emc during recirculation, this should be considered along with the cuckoo distributor and how they can work together.
-- Cuckoo distributor Patch link: https://mail.openvswitch.org/pipermail/ovs-dev/2017-June/334030.html
-- Skip EMC on recirculation Patch link: https://mail.openvswitch.org/pipermail/ovs-dev/2017-June/334293.html 
-- Slides link: Once slides are uploaded to public location Ian will add link to google doc.

- Multi-segment mbuf
-- Mark Kavanagh is currently working to enable multi-segment mbuf.
-- Patch to help enable the this work submitted by Michael Qui.
-- This is key to enabling Multi-segment mbuf, has been reviewed and tested by Mark Kavanagh but no others.
-- Request for this patch to be reviewed so that work can continue.
-- Patch link: https://mail.openvswitch.org/pipermail/ovs-dev/2017-June/334274.html 

- tunneling : Improving tunneling performance by avoiding the recirculation on datapath.
-- Sugesh Chandran has flagged that this requires specific feedback regarding what is required to upstream the patch.
-- Ongoing discussions on ML to enable this.
-- Request for feedback on this patch.
-- Patch Link: https://mail.openvswitch.org/pipermail/ovs-dev/2017-July/334854.html 

================
HARDWARE OFFLOAD
================

- Feedback given with regards to Napatech and Mellanox HWOL rfcs.
-- Current approach is HW is used to configured megaflows for OVS.
-- Both Napatech and Mellanox patches make use of RTE_FLOW api.
-- General feeling is the RTE_FLOW api is limited in scope, needs additional work to deal with more usecases.
-- Mellanox has work in progress to bypass circulation of vxlans.
-- Napatech Patch Link: https://mail.openvswitch.org/pipermail/ovs-dev/2017-June/334708.html 
-- Mellanox Patch link: https://mail.openvswitch.org/pipermail/ovs-dev/2017-July/334901.html 







More information about the dev mailing list