On 12 April 2017 at 15:57, Simon Horman <[email protected]> wrote: > On Wed, Apr 12, 2017 at 03:01:29PM -0300, Flavio Leitner wrote: >> On Sat, Apr 08, 2017 at 04:47:57PM -0400, Simon Horman wrote: >> > At Netdev 2.1 a meeting was held to discuss OvS offload. Minutes of the >> > discussion follow. I apologise in advance for any errors or omissions; >> > doubly for any errors in the attendee list. >> > >> > Topic: OVS Hardware Offload Using TC >> > Date: 7th April 2017 >> > Location: Netdev 2.1, Montreal >> > Attendees: Aaron Conole, Ben LaHaise, Eran Ben Elisha, Hannes Frederic >> > Sowa, >> > Jakub Kicinski, Jiri Pirko, Joe Stringer, John Fastabend, Nick Viljoen, >> > Rashid Khan, Rony Efraim, Simon Horman >> > >> > Joe raised 2 concerns: >> > >> > 1) How to enable users to understand whether offload is >> > successful and if not, why not? >> > >> > a) There is functionality in the v7[1] patchset to report which flows >> > are present in hardware. >> > >> > b) New error reporting infrastructure from the kernel is forthcoming It >> > should allow TC to provide more error information if a flow can't be >> > added to hardware. This could be made available to users - e.g. logged >> > - to allow them better understand the reason for the failure. >> > >> > 2) Maintenance burden falling on existing maintainers >> > >> > a) Simon offered to take some of the maintenance burden >> > immediately as he is already a committer. >> > >> > b) The aim is to ensure that in future there are other committers >> > who are interested in this feature. >> > >> > There was consensus that if the feature-set does not grow there should be >> > discussion of deprecating the HW offload support provided by [1]. >> > >> > Joe raised issue of whether OVS should probe hardware capabilities at >> > runtime. >> > John suggested this may be complex; potential combinatorial set is too >> > large. >> > >> > Rony then raised the increased complexity of using multiple NICs of >> > different types with different offload capabilities, this was tabled to a >> > later date. >> > >> > Joe has expressed a desire for more testing. There was a general agreement >> > to contribute tests. >> > >> > [1] [PATCH ovs V7 00/24] Introducing HW offload support for openvswitch >> >> Thanks for the minutes. I wasn't there so this helps to understand >> what has been discussed. >> >> Have you discussed how we are going to tag/document this feature? For >> instance, are we going to say this is "experimental"? > > My recollection is that important detail was not discussed at the meeting.
Correct, this did not come up during the meeting but it did come up a couple of times in other conversations at netdev. There seemed to be high level acknowledgement that it would be reasonable to document this feature in some way to indicate it's not ready for widespread consumption, so I figured that we can discuss the particulars of this further on the list. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
