Skip to contentSkip to conversation

netdev development stats for 6.6

ANN Jakub Kicinski 2 participants 3 messages Aug 29, 2023 – Sep 1, 2023

Subsystems (2)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

Hi!

General stats
-------------

The cycle started on June 27th and ended on August 28th, it was exactly the same length as the previous release cycle.

We have seen total of 16834 messages on the list (272 / day) which is 10% higher than last time (but very similar to the 6.4 cycle). The number of commits directly applied by netdev maintainers increased by 20% to 22 commits a day breaking the 3 release long streak of 18 / day. The increases seem a little surprising to me, TBH, after all this release covered a lot of the vacation season.

The number of participants in the ML discussions continues to oscillate around 770 individuals.

The patch review rate (Review/Ack tags) has recovered back to the 6.4 level and is respectively at 66% and 58%. Hopefully the 6.5 dip was temporary.

Rankings
--------

Top reviewers (thr): Top reviewers (msg):

   1 (   ) [44] Simon Horman            1 (   ) [59] Simon Horman        
   2 (   ) [33] Jakub Kicinski          2 (   ) [56] Jakub Kicinski      
   3 (   ) [13] Andrew Lunn             3 (   ) [30] Andrew Lunn         
   4 ( +9) [11] Leon Romanovsky         4 ( +9) [21] Leon Romanovsky     
   5 ( -1) [ 8] Paolo Abeni             5 (   ) [13] Russell King        
   6 ( -1) [ 6] Eric Dumazet            6 ( +5) [11] David Ahern         
   7 ( +4) [ 6] David Ahern             7 (+33) [10] Jesper Dangaard Brouer
   8 ( +6) [ 6] Ido Schimmel            8 ( +1) [10] Paolo Abeni         
   9 ( -3) [ 6] Russell King            9 ( -5) [ 9] Eric Dumazet        

10 ( +5) [ 5] Willem de Bruijn 10 ( +9) [ 9] Ido Schimmel
11 (+41) [ 5] Jesper Dangaard Brouer 11 ( +5) [ 8] Willem de Bruijn
12 (+24) [ 4] Alexander Lobakin 12 ( -2) [ 8] Michael S. Tsirkin

The top reviewer ranking is stable enough, with Simon topping the list (as well as the kernel-wide reviewer list, congrats!).
Ido returns to the top 12 by actively reviewing core routing as well as some "protocol driver" code. Jesper jumps into top 12 registering the highest move from his work reviewing the page pool code. Leon returns to #4 after a less active 6.5 cycle.

Top authors (thr): Top authors (msg):

   1 (***) [8] Yue Haibing              1 ( +2) [24] Tony Nguyen         
   2 ( -1) [7] Jakub Kicinski           2 (   ) [19] Saeed Mahameed      
   3 ( -1) [5] Eric Dumazet             3 ( +2) [17] Jakub Kicinski      
   4 ( -1) [5] Tony Nguyen              4 (+21) [17] Eric Dumazet        
   5 (***) [4] Jinjie Ruan              5 (***) [16] Qi Zheng            
   6 (***) [3] Suman Ghosh              6 (+25) [14] Jiri Pirko          
   7 ( -1) [3] Kuniyuki Iwashima        7 (***) [13] Hannes Reinecke     
   8 ( +2) [3] Daniel Golle             8 (+18) [11] Vladimir Oltean     
   9 ( +6) [3] Lin Ma                   9 (+15) [11] Dmitry Safonov      

10 (***) [3] Hangbin Liu 10 (+31) [10] Larysa Zaremba

Yue Haibing tops the raking of most threads started with the work of removing unused declarations from the kernel.
Jinjie Ruan authored patches removing of_match_ptr() annotations and "fixing" minor issues with error handling.

Qi Zheng posted the huge "refcount+RCU method to implement lockless slab shrink" series a few times, not really related to netdev. Hannes Reinecke worked on TLS and TLS + NVMe integration.

Company rankings
----------------

Top reviewers (thr): Top reviewers (msg):

   1 (   ) [44] Corigine                1 ( +1) [66] Meta                
   2 (   ) [37] Meta                    2 ( -1) [59] Corigine            
   3 (   ) [23] RedHat                  3 (   ) [51] RedHat              
   4 ( +3) [23] nVidia                  4 ( +2) [46] nVidia              
   5 ( -1) [17] Intel                   5 ( -1) [32] Intel               
   6 (   ) [14] Google                  6 ( -1) [30] Andrew Lunn         
   7 ( -2) [13] Andrew Lunn             7 (   ) [29] Google              

The company ranking is pretty stable, with only notable movement being nVidia's return to #4.

Top authors (thr): Top authors (msg):

   1 ( +5) [25] Huawei                  1 (   ) [86] Intel               
   2 ( -1) [19] Intel                   2 ( +1) [70] nVidia              
   3 ( -1) [17] RedHat                  3 ( -1) [52] RedHat              
   4 (   ) [12] Meta                    4 ( +6) [44] Huawei              
   5 ( -2) [11] nVidia                  5 (   ) [38] Meta                
   6 ( -1) [10] Google                  6 ( +7) [32] Google              
   7 ( +5) [ 9] Marvell                 7 ( +1) [25] AMD                 

Development vs reviewing scores
-------------------------------

Top scores (positive): Bottom scores (negative):

   1 (   ) [542] Corigine               1 (+31) [136] Huawei             
   2 (   ) [346] Meta                   2 (   ) [106] Intel              
   3 (   ) [188] Andrew Lunn            3 ( +7) [ 82] Bytedance          
   4 ( +1) [124] RedHat                 4 ( +4) [ 76] AMD                
   5 ( +1) [ 87] Linaro                 5 (***) [ 70] SUSE               
   6 ( +1) [ 82] Enfabrica              6 ( +7) [ 57] Marvell            

7 ( -3) [ 67] Google
8 ( +2) [ 58] Linux Foundation
9 (+12) [ 56] Broadcom
10 ( +4) [ 52] ARM
11 (***) [ 45] Microchip
12 ( +3) [ 41] nVidia

The semi-automated changes from Huawei have returned after a few releases of relative calm, putting the company firmly as #1 negative contributor from the reviewing perspective. This saves Intel from going to #1, but note that in absolute terms the "negative" side of the ranking got a lot worse. Previously first three positions had scores of 55, 42 and 40, so they would *not even make the top 6* now! SUSE is likely a temporary blip due to work from Hannes, but the other companies on the list should take a hard look at themselves.

Not much movement on the "good" side. With the exception of nVidia, the changes there seem to be mostly by posting volume not increase in reviews.

Thanks to everyone who contributed code to networking trees in this release cycle, and huge thanks to everyone reviewing!

--

No major changes to the stats generation process in this release, I focused on minor improvements to parsing and matching people to companies.

Code: https://github.com/kuba-moo/ml-stat
author maintainer
#1 lore raw

A minor follow up / question.

Would it be helpful to generate these kind of stats scoped to particular companies?

Very few people end up in the top 12. It may not be entirely trivial to run these scripts locally because people use their non-company addresses at times. And because of GDPR and data privacy concerns I don't want to upload raw results to GitHub or some such :(

The best idea I could come up with was to post the stats to a per-company ML. Intel has the "wired" list which is a perfect fit. That way the result is accessible to developers, and they can refer to it in the "performance review" or whenever they want to get credit for upstream reviews? :)

Does that sound sensible?

Please don't hesitate to LMK either on or off list. Particularly if you think this is a bad idea.

And LMK if any other company wants similar stats.

#2 lore raw
Quoted 16 lines · Jakub Kicinski

A minor follow up / question.

Would it be helpful to generate these kind of stats scoped to particular companies?

Very few people end up in the top 12. It may not be entirely trivial to run these scripts locally because people use their non-company addresses at times. And because of GDPR and data privacy concerns I don't want to upload raw results to GitHub or some such :(

The best idea I could come up with was to post the stats to a per-company ML. Intel has the "wired" list which is a perfect fit. That way the result is accessible to developers, and they can refer to it in the "performance review" or whenever they want to get credit for upstream reviews? :)

Quoted 6 lines

Does that sound sensible?

Please don't hesitate to LMK either on or off list. Particularly if you think this is a bad idea.

And LMK if any other company wants similar stats.

We got ours for the cycle, thank you! I think this is a great idea.