netdev development stats for 6.1?
Subsystems (2)
Jakub Kicinski
maintainerHi!
For a while now I had been curious if we can squeeze any interesting stats from the ML traffic. In particular I was curious "who is helping", who is reviewing the most patches (but based on the emails sent not just review tags).
I quickly wrote a script to scan emails sent to netdev since 5.19 was tagged (~14k) and count any message which has subject starting with '[' as a patch and anything else as a comment/review. It's not very scientific but the result for the most part matches my expectations.
A disclaimer first - this methodology puts me ahead because I send a lot of emails. Most of them are not reviews, so ignore me.
Second question to address upfront is whether publishing stats is useful or mostly risks people treating participation as a competition and trying to game the system? Hard to say, but if even a single person can point to these stats to help justify more time spent reviewing to their management - it's worth it.
That said feedback is very welcome, public or private.
The stats are by number of threads and number of messages.
Top 10 reviewers (thr): Top 10 reviewers (msg):
1. [320] Jakub Kicinski 1. [538] Jakub Kicinski 2. [134] Andrew Lunn 2. [263] Andrew Lunn 3. [ 51] Krzysztof Kozlowski 3. [122] Krzysztof Kozlowski 4. [ 51] Paolo Abeni 4. [ 80] Rob Herring 5. [ 47] Eric Dumazet 5. [ 78] Eric Dumazet 6. [ 46] Rob Herring 6. [ 70] Paolo Abeni 7. [ 35] Florian Fainelli 7. [ 65] Vladimir Oltean 8. [ 35] Kalle Valo 8. [ 58] Ido Schimmel 9. [ 32] David Ahern 9. [ 58] Michael S. Tsirkin
10. [ 31] Vladimir Oltean 10. [ 57] Russell King
These seem to make sense, but the volume-centric view shows. Note that the numbers are very close so the exact order is of little importance. The names should be familiar to everyone, I hope :)
Top 10 authors (thr): Top 10 authors (msg):
1. [ 84] Zhengchao Shao 1. [287] Zhengchao Shao 2. [ 52] Vladimir Oltean 2. [232] Vladimir Oltean 3. [ 43] Jakub Kicinski 3. [166] Saeed Mahameed 4. [ 28] Tony Nguyen 4. [156] Kuniyuki Iwashima 5. [ 28] [email protected] 5. [134] Sean Anderson 6. [ 23] Stephen Rothwell 6. [122] Oleksij Rempel 7. [ 23] Hangbin Liu 7. [106] Tony Nguyen 8. [ 20] Wolfram Sang 8. [ 93] Mattias Forsblad 9. [ 20] Kuniyuki Iwashima 9. [ 93] Jian Shen
10. [ 20] Jiri Pirko 10. [ 86] Jakub Kicinski
Here Stephen is probably by accident as I was counting his merge resolutions as patches.
What is clear tho (with the notable exception of Vladimir)
- most of the authors are not making the top reviewer list :(
And here is the part that I was most curious about. Calculate a "score" which is roughly:
10 * reviews - 3 * authorship,
to see who is a "good citizen":
Top 10 scores (positive): Top 10 scores (negative):
1. [4102] Jakub Kicinski 1. [397] Zhengchao Shao 2. [1848] Andrew Lunn 2. [116] Kuniyuki Iwashima 3. [737] Krzysztof Kozlowski 3. [105] [email protected] 4. [620] Paolo Abeni 4. [ 93] Mattias Forsblad 5. [611] Rob Herring 5. [ 82] Yang Yingliang 6. [588] Eric Dumazet 6. [ 82] Sean Anderson 7. [429] Florian Fainelli 7. [ 77] Daniel Lezcano 8. [418] Kalle Valo 8. [ 68] Stephen Rothwell 9. [406] David Ahern 9. [ 67] Arun Ramadoss
10. [344] Russell King 10. [ 64] Wang Yufen
Now looking at companies.
[Using my very rough mapping of people to company based on email domain and manual mapping for major contributors]
Top 7 reviewers (thr): Top 7 reviewers (msg):
1. [369] Meta 1. [640] Meta 2. [139] Intel 2. [306] RedHat 3. [134] Andrew Lunn 3. [263] Andrew Lunn 4. [127] RedHat 4. [243] Intel 5. [ 80] nVidia 5. [193] nVidia 6. [ 71] Google 6. [134] Linaro 7. [ 61] Linaro 7. [121] Google
Top 8 authors (thr): Top 7 authors (msg):
1. [207] Huawei 1. [640] Huawei 2. [103] nVidia 2. [496] nVidia 3. [ 96] Intel 3. [342] Intel 4. [ 94] RedHat 4. [332] RedHat 5. [ 75] Google 5. [263] NXP 6. [ 60] Microchip 6. [170] Linaro 7. [ 59] NXP 7. [157] Amazon
8. [ 51] Meta
Top 12 scores (positive): Top 12 scores (negative):
1. [4763] Meta 1. [887] Huawei 2. [1848] Andrew Lunn 2. [145] Microchip 3. [1432] RedHat 3. [105] ZTE 4. [1415] Intel 4. [ 95] Amazon 5. [ 680] Linaro 5. [ 93] Mattias Forsblad 6. [ 652] Google 6. [ 68] Stephen Rothwell 7. [ 627] nVidia 7. [ 59] Wolfram Sang 8. [ 609] Rob Herring 8. [ 57] [email protected] 9. [ 429] Florian Fainelli 9. [ 56] Arınç ÜNAL 10. [ 418] Kalle Valo 10. [ 53] Sean Anderson 11. [ 368] Russell King 11. [ 48] Maxime Chevallier 12. [ 356] David Ahern 12. [ 46] Jianguo Zhang
The bot operators top the list of "bad citizens" as they do not contribute to the review process. Microchip and Amazon also seem to send a lot more code than they help to review.
Huge *thank you* to all the reviewers!
Quoted 15 lines · Jakub Kicinski
Hi!
For a while now I had been curious if we can squeeze any interesting stats from the ML traffic. In particular I was curious "who is helping", who is reviewing the most patches (but based on the emails sent not just review tags).
I quickly wrote a script to scan emails sent to netdev since 5.19 was tagged (~14k) and count any message which has subject starting with '[' as a patch and anything else as a comment/review. It's not very scientific but the result for the most part matches my expectations.
A disclaimer first - this methodology puts me ahead because I send a lot of emails. Most of them are not reviews, so ignore me.
Quoted 7 lines
Second question to address upfront is whether publishing stats is useful or mostly risks people treating participation as a competition and trying to game the system? Hard to say, but if even a single person can point to these stats to help justify more time spent reviewing to their management - it's worth it.
That said feedback is very welcome, public or private.
I think that it is right initiative which will make netdev community stronger and wider.
As for the feedback, which express my personal view as an outsider in netdev.
I think that more clear goals for that statistics can help to purify which tables are actually needed as I'm sure that not all are needed.
My goals are:
1. See that load spreads more equity. It will indirectly cause to spread of the knowledge. The most active reviewers are the most knowledgeable developers too.
2. Push companies to participate in code maintenance (review) and not only enjoy from free rides.
Thanks
Quoted 122 lines · Jakub Kicinski
Hi!
For a while now I had been curious if we can squeeze any interesting stats from the ML traffic. In particular I was curious "who is helping", who is reviewing the most patches (but based on the emails sent not just review tags).
I quickly wrote a script to scan emails sent to netdev since 5.19 was tagged (~14k) and count any message which has subject starting with '[' as a patch and anything else as a comment/review. It's not very scientific but the result for the most part matches my expectations.
A disclaimer first - this methodology puts me ahead because I send a lot of emails. Most of them are not reviews, so ignore me.
Second question to address upfront is whether publishing stats is useful or mostly risks people treating participation as a competition and trying to game the system? Hard to say, but if even a single person can point to these stats to help justify more time spent reviewing to their management - it's worth it.
That said feedback is very welcome, public or private.
The stats are by number of threads and number of messages.
Top 10 reviewers (thr): Top 10 reviewers (msg):
1. [320] Jakub Kicinski 1. [538] Jakub Kicinski 2. [134] Andrew Lunn 2. [263] Andrew Lunn 3. [ 51] Krzysztof Kozlowski 3. [122] Krzysztof Kozlowski 4. [ 51] Paolo Abeni 4. [ 80] Rob Herring 5. [ 47] Eric Dumazet 5. [ 78] Eric Dumazet 6. [ 46] Rob Herring 6. [ 70] Paolo Abeni 7. [ 35] Florian Fainelli 7. [ 65] Vladimir Oltean 8. [ 35] Kalle Valo 8. [ 58] Ido Schimmel 9. [ 32] David Ahern 9. [ 58] Michael S. Tsirkin10. [ 31] Vladimir Oltean 10. [ 57] Russell King
These seem to make sense, but the volume-centric view shows. Note that the numbers are very close so the exact order is of little importance. The names should be familiar to everyone, I hope :)
Top 10 authors (thr): Top 10 authors (msg):
1. [ 84] Zhengchao Shao 1. [287] Zhengchao Shao 2. [ 52] Vladimir Oltean 2. [232] Vladimir Oltean 3. [ 43] Jakub Kicinski 3. [166] Saeed Mahameed 4. [ 28] Tony Nguyen 4. [156] Kuniyuki Iwashima 5. [ 28] [email protected] 5. [134] Sean Anderson 6. [ 23] Stephen Rothwell 6. [122] Oleksij Rempel 7. [ 23] Hangbin Liu 7. [106] Tony Nguyen 8. [ 20] Wolfram Sang 8. [ 93] Mattias Forsblad 9. [ 20] Kuniyuki Iwashima 9. [ 93] Jian Shen10. [ 20] Jiri Pirko 10. [ 86] Jakub Kicinski
Here Stephen is probably by accident as I was counting his merge resolutions as patches.
What is clear tho (with the notable exception of Vladimir)
- most of the authors are not making the top reviewer list :(And here is the part that I was most curious about. Calculate a "score" which is roughly:
10 * reviews - 3 * authorship,
to see who is a "good citizen":Top 10 scores (positive): Top 10 scores (negative):
1. [4102] Jakub Kicinski 1. [397] Zhengchao Shao 2. [1848] Andrew Lunn 2. [116] Kuniyuki Iwashima 3. [737] Krzysztof Kozlowski 3. [105] [email protected] 4. [620] Paolo Abeni 4. [ 93] Mattias Forsblad 5. [611] Rob Herring 5. [ 82] Yang Yingliang 6. [588] Eric Dumazet 6. [ 82] Sean Anderson 7. [429] Florian Fainelli 7. [ 77] Daniel Lezcano 8. [418] Kalle Valo 8. [ 68] Stephen Rothwell 9. [406] David Ahern 9. [ 67] Arun Ramadoss10. [344] Russell King 10. [ 64] Wang Yufen
Now looking at companies.
[Using my very rough mapping of people to company based on email domain and manual mapping for major contributors]
Top 7 reviewers (thr): Top 7 reviewers (msg):
1. [369] Meta 1. [640] Meta 2. [139] Intel 2. [306] RedHat 3. [134] Andrew Lunn 3. [263] Andrew Lunn 4. [127] RedHat 4. [243] Intel 5. [ 80] nVidia 5. [193] nVidia 6. [ 71] Google 6. [134] Linaro 7. [ 61] Linaro 7. [121] GoogleTop 8 authors (thr): Top 7 authors (msg):
1. [207] Huawei 1. [640] Huawei 2. [103] nVidia 2. [496] nVidia 3. [ 96] Intel 3. [342] Intel 4. [ 94] RedHat 4. [332] RedHat 5. [ 75] Google 5. [263] NXP 6. [ 60] Microchip 6. [170] Linaro 7. [ 59] NXP 7. [157] Amazon8. [ 51] Meta
Top 12 scores (positive): Top 12 scores (negative):
1. [4763] Meta 1. [887] Huawei 2. [1848] Andrew Lunn 2. [145] Microchip 3. [1432] RedHat 3. [105] ZTE 4. [1415] Intel 4. [ 95] Amazon 5. [ 680] Linaro 5. [ 93] Mattias Forsblad 6. [ 652] Google 6. [ 68] Stephen Rothwell 7. [ 627] nVidia 7. [ 59] Wolfram Sang 8. [ 609] Rob Herring 8. [ 57] [email protected] 9. [ 429] Florian Fainelli 9. [ 56] Arınç ÜNAL 10. [ 418] Kalle Valo 10. [ 53] Sean Anderson 11. [ 368] Russell King 11. [ 48] Maxime Chevallier 12. [ 356] David Ahern 12. [ 46] Jianguo Zhang
Quoted 5 lines
The bot operators top the list of "bad citizens" as they do not contribute to the review process. Microchip and Amazon also seem to send a lot more code than they help to review.
Huge *thank you* to all the reviewers!
One statistic that I would be curious to have is the ratio of features vs. fixes and who fixes bugs from others versus fixing their own bugs.
Interesting stats and thanks for putting those together!
Signature · 1 line
Florian
Quoted 122 lines · Jakub Kicinski
Hi!
For a while now I had been curious if we can squeeze any interesting stats from the ML traffic. In particular I was curious "who is helping", who is reviewing the most patches (but based on the emails sent not just review tags).
I quickly wrote a script to scan emails sent to netdev since 5.19 was tagged (~14k) and count any message which has subject starting with '[' as a patch and anything else as a comment/review. It's not very scientific but the result for the most part matches my expectations.
A disclaimer first - this methodology puts me ahead because I send a lot of emails. Most of them are not reviews, so ignore me.
Second question to address upfront is whether publishing stats is useful or mostly risks people treating participation as a competition and trying to game the system? Hard to say, but if even a single person can point to these stats to help justify more time spent reviewing to their management - it's worth it.
That said feedback is very welcome, public or private.
The stats are by number of threads and number of messages.
Top 10 reviewers (thr): Top 10 reviewers (msg):
1. [320] Jakub Kicinski 1. [538] Jakub Kicinski 2. [134] Andrew Lunn 2. [263] Andrew Lunn 3. [ 51] Krzysztof Kozlowski 3. [122] Krzysztof Kozlowski 4. [ 51] Paolo Abeni 4. [ 80] Rob Herring 5. [ 47] Eric Dumazet 5. [ 78] Eric Dumazet 6. [ 46] Rob Herring 6. [ 70] Paolo Abeni 7. [ 35] Florian Fainelli 7. [ 65] Vladimir Oltean 8. [ 35] Kalle Valo 8. [ 58] Ido Schimmel 9. [ 32] David Ahern 9. [ 58] Michael S. Tsirkin10. [ 31] Vladimir Oltean 10. [ 57] Russell King
These seem to make sense, but the volume-centric view shows. Note that the numbers are very close so the exact order is of little importance. The names should be familiar to everyone, I hope :)
Top 10 authors (thr): Top 10 authors (msg):
1. [ 84] Zhengchao Shao 1. [287] Zhengchao Shao 2. [ 52] Vladimir Oltean 2. [232] Vladimir Oltean 3. [ 43] Jakub Kicinski 3. [166] Saeed Mahameed 4. [ 28] Tony Nguyen 4. [156] Kuniyuki Iwashima 5. [ 28] [email protected] 5. [134] Sean Anderson 6. [ 23] Stephen Rothwell 6. [122] Oleksij Rempel 7. [ 23] Hangbin Liu 7. [106] Tony Nguyen 8. [ 20] Wolfram Sang 8. [ 93] Mattias Forsblad 9. [ 20] Kuniyuki Iwashima 9. [ 93] Jian Shen10. [ 20] Jiri Pirko 10. [ 86] Jakub Kicinski
Here Stephen is probably by accident as I was counting his merge resolutions as patches.
What is clear tho (with the notable exception of Vladimir)
- most of the authors are not making the top reviewer list :(And here is the part that I was most curious about. Calculate a "score" which is roughly:
10 * reviews - 3 * authorship,
to see who is a "good citizen":Top 10 scores (positive): Top 10 scores (negative):
1. [4102] Jakub Kicinski 1. [397] Zhengchao Shao 2. [1848] Andrew Lunn 2. [116] Kuniyuki Iwashima 3. [737] Krzysztof Kozlowski 3. [105] [email protected] 4. [620] Paolo Abeni 4. [ 93] Mattias Forsblad 5. [611] Rob Herring 5. [ 82] Yang Yingliang 6. [588] Eric Dumazet 6. [ 82] Sean Anderson 7. [429] Florian Fainelli 7. [ 77] Daniel Lezcano 8. [418] Kalle Valo 8. [ 68] Stephen Rothwell 9. [406] David Ahern 9. [ 67] Arun Ramadoss10. [344] Russell King 10. [ 64] Wang Yufen
Now looking at companies.
[Using my very rough mapping of people to company based on email domain and manual mapping for major contributors]
Top 7 reviewers (thr): Top 7 reviewers (msg):
1. [369] Meta 1. [640] Meta 2. [139] Intel 2. [306] RedHat 3. [134] Andrew Lunn 3. [263] Andrew Lunn 4. [127] RedHat 4. [243] Intel 5. [ 80] nVidia 5. [193] nVidia 6. [ 71] Google 6. [134] Linaro 7. [ 61] Linaro 7. [121] GoogleTop 8 authors (thr): Top 7 authors (msg):
1. [207] Huawei 1. [640] Huawei 2. [103] nVidia 2. [496] nVidia 3. [ 96] Intel 3. [342] Intel 4. [ 94] RedHat 4. [332] RedHat 5. [ 75] Google 5. [263] NXP 6. [ 60] Microchip 6. [170] Linaro 7. [ 59] NXP 7. [157] Amazon8. [ 51] Meta
Top 12 scores (positive): Top 12 scores (negative):
1. [4763] Meta 1. [887] Huawei 2. [1848] Andrew Lunn 2. [145] Microchip 3. [1432] RedHat 3. [105] ZTE 4. [1415] Intel 4. [ 95] Amazon 5. [ 680] Linaro 5. [ 93] Mattias Forsblad 6. [ 652] Google 6. [ 68] Stephen Rothwell 7. [ 627] nVidia 7. [ 59] Wolfram Sang 8. [ 609] Rob Herring 8. [ 57] [email protected] 9. [ 429] Florian Fainelli 9. [ 56] Arınç ÜNAL 10. [ 418] Kalle Valo 10. [ 53] Sean Anderson 11. [ 368] Russell King 11. [ 48] Maxime Chevallier 12. [ 356] David Ahern 12. [ 46] Jianguo Zhang
Quoted 5 lines
The bot operators top the list of "bad citizens" as they do not contribute to the review process. Microchip and Amazon also seem to send a lot more code than they help to review.
Huge *thank you* to all the reviewers!
Well, it seems I've made some of the "negative" lists. In my defense,
- I had to make several revisions to the series I was working on. 134 sent patches turned into only 36 commits (so far). From what I've been told, it seems like you're supposed to resend the series after a week or so after making changes in response to feedback. I would prefer to resend less often, only after each patch without a RB has gotten feedback, but that usually doesn't happen for whatever reason.
- I don't think I'm familiar enough with the net subsystem to review most patches. I think I could review some phylink stuff, but that's about it. This subsystem is very intimidating. It often takes me a lot of effort to determine the correct thing to do in my own patches, much less someone else's patches for unfamiliar hardware.
- I *do* review patches... on other projects. I'm fairly active on the U-Boot mailing lists, where I review patches for the clock subsystem. Of course, I do that in my free time. I try to upstream the patches I write, but that's pretty unusual for my company. There's no allocated time for "giving back," and most of my peers don't see the value in it.
I think the approach taken here is a bit reductive, and not too holistic. That said, I do appreciate the netdev reviewers a lot. Submitting patches here is much nicer than in some other subsystems.
--Sean
Jakub Kicinski
maintainerQuoted 1 line · On Tue, 4 Oct 2022 21:27:21 -0700 Jakub Kicinski wrote:
That said feedback is very welcome, public or private.
Thanks a lot to all those who shared their thoughts, I had been mulling over all the good points that had been made. It's not an easy problem. I reckon next time around I'll share the script and perhaps more folks can try their hand at extracting the information?