2458 days of Linux development - some stats and thoughts

2458 days of Linux development - some stats and thoughts

[Jan III Sobieski]

[-- Attachment #1: Type: text/plain, Size: 952 bytes --]

Hi,

I'm sending some stats from 2.6 kernel development process in the attachment.

It seems to me that the kernel development is moving in really good direction:
- stable trees were a good invention
- long supported stable trees were even a better invention :)
- most changes are tested before inclusion in the tree next
- development of new drivers in the staging was a very good idea
- regression tracking works better and better
- latest Linus strict stabilization policy seems to be the next great milestone

What's next? Tracking changes in kernel performance and eliminating
performance regressions would be IMHO a great feature of Linux
development process :)

Phoronix is already doing such tests, but they do this to write
another "oh my god, this slowed down" article.

Perhaps the inclusion of Phoronix Test Suite to Autotest would be a
good idea? Once the automatic tests were made on autotest.kernel.org

Regards,
J

-- 
Jan III Sobieski

[-- Attachment #2: stats.ods --]
[-- Type: application/vnd.oasis.opendocument.spreadsheet, Size: 18809 bytes --]

Re: 2458 days of Linux development - some stats and thoughts

[Jan III Sobieski]

[-- Attachment #1: Type: text/plain, Size: 75 bytes --]

I added -next coverage statistics for a few kernels.

-- 
Jan III Sobieski

[-- Attachment #2: stats.ods --]
[-- Type: application/vnd.oasis.opendocument.spreadsheet, Size: 19652 bytes --]

Re: 2458 days of Linux development - some stats and thoughts

[Jan III Sobieski]

[-- Attachment #1: Type: text/plain, Size: 209 bytes --]

I added regressions stats - total number of regressions for each
kernel and number of regressions not fixed yet. It is interesting that
the number of not fixed regressions is really low.

-- 
Jan III Sobieski

[-- Attachment #2: stats.ods --]
[-- Type: application/vnd.oasis.opendocument.spreadsheet, Size: 19845 bytes --]

Re: 2458 days of Linux development - some stats and thoughts

[Florian Mickler]

On Sat, 11 Sep 2010 00:04:18 +0200
Jan III Sobieski <jan3sobi3ski@gmail.com> wrote:

> I added regressions stats - total number of regressions for each
> kernel and number of regressions not fixed yet. It is interesting that
> the number of not fixed regressions is really low.
> 

What should I do with these numbers, when I don't know
how they are created? What do they signify? What is their relevance?
What uncertainty is connected to these numbers? What follows? Why are
they created? How are they related? 

Cheers,
Flo

Re: 2458 days of Linux development - some stats and thoughts

[Jan III Sobieski]

2010/9/11 Florian Mickler <florian@mickler.org>:
> On Sat, 11 Sep 2010 00:04:18 +0200
> Jan III Sobieski <jan3sobi3ski@gmail.com> wrote:
>
>> I added regressions stats - total number of regressions for each
>> kernel and number of regressions not fixed yet. It is interesting that
>> the number of not fixed regressions is really low.
>>
>
> What should I do with these numbers, when I don't know
> how they are created? What do they signify? What is their relevance?
> What uncertainty is connected to these numbers? What follows? Why are
> they created? How are they related?

This is just simple stat based on diffs. For example - 2.6.32


32	1096601	534042	315020	228772	781581	305270	1086851	543792	1630643	85
 	8	882149	463400	1345549	82,52%	160	28


32 - version
1096601 - insertions for whole 2.6.32 patch
534042 - deletions for whole 2.6.32 patch
315020 - insertions in drivers/staging dir
228772 - deletions in drivers/staging dir
781581 - insertions in base (without staging - most people doesn't use
it and doesn't care about this code - it's just a noise in diff's)
305270 - deletions in base
1086851 - sum of base insertions and deletions
543792 - sum of staging insertions and deletions
1630643 - sum of insertions and deletions in whole patch
85 - how many days took to develop this kernel
8 - how many rc's was produced
882149 - insertions from latest -next tree for 2.6.32
463400 - deletions from latest -next tree for 2.6.32
1345549 - sum of above for latest -next tree for 2.6.32
82,52% - how many changed lines was tested in -next tree
160	- total number of regressions tracked for 2.6.32
28 - regressions that are not fixed yet

I see a trend in the latest kernels (since 2.6.32) to minimize sum of
changes in base (without drivers/staging) - average is less than
1086851 lines changed and regressions dropped down bellow 160 for each
release. From 2.6.24 to 2.6.31 there were a lot more changes in base
and total number of regressions for each version was above 160 (with
one exception).

You can also see some dependency between number of regressions and
percent of changes coveraged in -next tree.

But this is just random stats - you can see whatever you want ;)

>
> Cheers,
> Flo
>

-- 
Jan III Sobieski