[ovs-dev] [PATCH v11 0/5] dpcls func ptrs & optimizations

Ilya Maximets i.maximets at samsung.com
Wed Jul 17 10:18:41 UTC 2019


On 17.07.2019 0:06, Ian Stokes wrote:
> On 7/15/2019 5:36 PM, Harry van Haaren wrote:
>> Hey Folks,
>>
>> Here a v11 of the DPCLS Function Pointer patchset, as has been
>> presented at OVS Conf in Nov '18, and discussed on the ML since then.
>> I'm aware of the soft-freeze for 2.12, I feel this patchset has had
>> enough reviews/versions/testing to be merged in 2.12.
>>
>> Thanks Ilya and Ian for review comments on v10, they should all be addressed
>> in this v11. Patchset details below, summary of v11 changes as follows:
>> - Reworked to use hash_add_words64(), unfortunatly hash_words64_inline() did
>>    not provide the same has due to it calling hash_finish() with a different
>>    "final" value. Refactored to re-use as much as we can of the existing code.
>> - Reworked specialized functions to use MACROs as suggested in review. This
>>    makes the specialization of functions much more tidy - good suggestion!
>> - Reworked lots of little fixes, Captials and stops.
>> - Added NEWS entry in "Userspace Datapath" section on DPCLS function pointers
>>    and how specialization of subtables minfilows gains search performance.
>> - See per patch --- v11 notes for details :)
>>
>> Regards, -Harry
> 
> Thanks for working on the v11 Harry.
> 
> Patches 1-4 in the series looks ok to me. I have some minor comments on patch 5.
> 
> I've ran it through the usual validation (travis, appveyor, read the docs etc). All passing without issue.
> 
> I'm just running some performance tests now again with vsperf.
> 
> For rfc2544 0% traffic loss for ovs flow scalability (phy2phy_scalability) I'm seeing a 10% increase in performance for 64 byte packets and in the case of scalability where loss is allowed (phy2phy_scalability_cont) then I see an 8% increase in performance on these baselines.
> 
> @Ilya, do you have input on these patches or has Harry addressed your concerns in the latest revision?

I've sent some comments for the patches.

BTW, I'm planning to test the performance of generic (non-optimized)
implementation today to compare with the previous implementation.

Best regards, Ilya Maximets.


More information about the dev mailing list