Re: codel: split into multiple files

[Jens Axboe <axboe@fb.com>]

On 04/26/2016 06:36 AM, Michal Kazior wrote:
> On 26 April 2016 at 08:43, Sedat Dilek <sedat.dilek@gmail.com> wrote:
>> On 4/26/16, Michal Kazior <michal.kazior@tieto.com> wrote:
>>> On 26 April 2016 at 08:09, Sedat Dilek <sedat.dilek@gmail.com> wrote:
>>>> Hi,
>>>>
>>>> I had a very quick view on net-next.git#master (up to commit
>>>> fab7b629a82da1b59620470d13152aff975239f6).
>>>>
>>>> Commit in [1] aka "codel: split into multiple files" removed codel.h
>>>> but [2] and [3] have relicts to it.
>>>> Forgot to remove?
>>>
>>> codel.h was not removed. diffstat for codel.h is all red which I
>>> presume is why you thought of it as removed, see:
>>>
>>> http://git.kernel.org/cgit/linux/kernel/git/davem/net-next.git/tree/include/net/codel.h?id=d068ca2ae2e614b9a418fb3b5f1fd4cf996ff032
>>>
>>
>> [ CC Jens ]
>>
>> OK.
>> So what are the plans in the future?
>> Keep a "generic" codel.h (compatibility reasons?) for net or is it your split?
>
> I'm interested in re-using codel in mac80211 for wireless. cfg80211
> drivers may want to do that as well later. Even vendor drivers could
> start to use it (I can dream :).
>
> I plan to re-spin my patches soonish re-based on the new codel.h/fq.h
> approach. There's quite a few spins already[1].
>
>
>> AFAICS I have seen a codel-implementation in block.git#wb-buf-throttle.
>> Does it make sense to have a more "super-generic" codel.h for re-use
>> (not only for net and block)?
>> Just a thought.
>
> Oh, I'm not really familiar with block and problems around it but it
> sounds reasonable and interesting. It doesn't look like it blatantly
> copies codel though (I did that in my initial mac80211 patches with
> some adjustments, you can check that in the link[1] which you can
> lookup via my patchset's cover letter[2]; I've based off of codel5[3]
> back then).

The block version is an adaptation, I guess you can say it pays homage 
to CoDel. But there are a sufficient amount of differences between 
networking and storage that I don't think a fully generic version is 
really feasible. My favorite thing to bring up is the fact that we don't 
have the luxury of dropping packets on the storage side...

-- 
Jens Axboe