[ovs-dev] [PATCH ovs-dev v3 1/4] dpif-netdev: Expand the meters supported number.

Ilya Maximets i.maximets at ovn.org
Mon Apr 26 17:11:30 UTC 2021


On 4/26/21 2:04 PM, Tonghao Zhang wrote:
> On Thu, Apr 15, 2021 at 1:04 AM Ilya Maximets <i.maximets at ovn.org> wrote:
>>
>> On 4/14/21 3:55 AM, Tonghao Zhang wrote:
>>> On Thu, Feb 25, 2021 at 2:17 AM Ilya Maximets <i.maximets at ovn.org> wrote:
>>>>
>>>> On 2/24/21 1:31 PM, Tonghao Zhang wrote:
>>>>> Now this patch version is v3. and stay a long time. Any maintainer
>>>>> will continue to review this patch ?  Thanks!
>>>>
>>>> Sorry for long dalays.  I have it on my TODO list for this week along
>>>> with the overflow fix v2.
>>> Hi Ilya, Ben
>>> Do you have comments on
>>> 1/4 dpif-netdev: Expand the meters supported number ?
>>
>> Hi.
>>
>> The main problem I have with the patch is the way how new data structure
>> introduced.  And I'm not really comfortable adding a new RCU-based structure
>> without full set of actual unit tests including tests of it's parallel
>> operation.
>>
>> You mentioned before that cmap has 1% overhead in compare with array.  Maybe
>> it's not that bad?  I mean, I'd sacrifice 1% of performance for the
>> simplicity of the implementation.  Maybe we can even optimize something in
>> you implementation or save this 1% at some different place.
>>
>> WDYT?
> The original patches implemented as cmap, and add a cache using meters array:
> http://patchwork.ozlabs.org/project/openvswitch/patch/1584180230-89020-1-git-send-email-xiangxia.m.yue@gmail.com/

Thanks for the pointer.  I prefer this implementation with cmap
and I see some possible small performance improvements that could
be done in the patch above.  I'll review it.

> http://patchwork.ozlabs.org/project/openvswitch/patch/1584180230-89020-2-git-send-email-xiangxia.m.yue@gmail.com/

This patch with a special cache seems too complex for the performance
benefit it has.  I'd rather not have it and try to optimize simple
cmap implementation instead.

> 
> Ben suggested I use the idpool, the kernel do this in the same way.
> Ben  what do you think ?
> The two versions look good to me.
> 
>> Best regards, Ilya Maximets.


More information about the dev mailing list