On Wed, 18 Jun 2025, Emmanuel Nyarko wrote:
Hi all,
I have already spoken to a few developers about this and now want to make
everyone aware.
The current state of ALTQ:
ALTQ code is very large with so maany queueing disciplines.
The state of the art per my research is Fair Queuing CoDel FQ-Codel.(i did the
work in making CoDel work in netbsd).
The previous disciplines are not well maintained and adding another discipline
makes it worse(maintainability).
i want to update it to FQ-CoDel and bring it in now so we use it as our primary
and maybe only active queue managements.
How do i want to bring it in ?
1. Remove all the existing disciplines (openBSD has taken this step already)
2. Use fq-codel which is easy to configure with fewer parameters and proves to
be very effecitve in mitigating bufferbloat
3. modularize it (so we don’t have to recompile kernel anytime one wants to use
it. as opposed to the current)
4. find a way to write tests for it
5. update the docs.
Modularizing:
this will also help for easy integration into NPF.
all points of contacts to ALTQ will be changed to agree with the decision eg:
in if_inittialise( when a new network device is connected) and also when
detached for each network interface.
Hence, we can have a modular, and smaller quality of service codebase to use
and maintain.
Please note that we also need to modularize the ALTQ code itself,
preferably prior to modularizing the queueing disciplines. I looked
at it once a while ago, but did not proceed due to lack of knowledge
re ALTQ itself.
Good luck!
+---------------------+--------------------------+----------------------+
| Paul Goyette (.sig) | PGP Key fingerprint: | E-mail addresses: |
| (Retired) | 1B11 1849 721C 56C8 F63A | p...@whooppee.com |
| Software Developer | 6E2E 05FD 15CE 9F2D 5102 | pgoye...@netbsd.org |
| & Network Engineer | | pgoyett...@gmail.com |
+---------------------+--------------------------+----------------------+
