Introduce generic slowstart algorithms for congestion control

[parvit]

Description:

This PR introduces the new functionality of 'slowstart algorithms' which are a step in congestion control algorithms that govern the initial handling of the congestion window.

This is to try to reduce initial packet loss during a connection and better estimate a correct congestion window.

The implemented algorithms are the following:

• rfc2001 : The default behavior for CCAs described in RFC2001 and implemented previously in quicly that doubles the congestion window until a loss is detected, at which point the slowstart phase is exited.

• disabled : For some congestion control algorithms (namely pico), rfc2001 behavior is not coherent and so an empty type of slowstart logic is provided.

• search : The SEARCH algorithm described in:

  • Amber Cronin, Maryam Ataei Kachooei, Jae Chung, Feng Li, Benjamin Peters, and Mark Claypool.
    Improving QUIC Slow Start Behavior in Wireless Networks with SEARCH, In Proceedings of the IEEE
    Local and Metropolitan Area Conference (LANMAN), Boston, MA, USA, July 2024.

  • Maryam Ataei Kachooei, Jae Chung, Feng Li, Benjamin Peters, Josh Chung, and
    Mark Claypool. Improving TCP Slow Start Performance in Wireless Networks with
    SEARCH, In Proceedings of the World of Wireless, Mobile and Multimedia Networks
    (WoWMoM), Perth, Australia June 2024.

  This algorithm tracks delivered bytes in 'bins' and averages their sum over time until a certain threshold is achieved, at that point the congestion window is updated and slowstart phase exited.

The three implementations are provided via a new struct quicly_ss_type_t, similar to quicly_cc_type_t and allows to select the current slowstart algorithm being used, by changing the pointer added in quicly context to the requested slowstart type.

If this is context field is left empty than no change is made and the default previous behavior is still used.

Please review and send feedback.

[kazuho]

Wow. Thank you for the pull request. We will take a look.

[parvit]

Wow. Thank you for the pull request. We will take a look.

Hi, did you have a chance to review this PR? thank you.

[vmfoxworthy]

Is this still under review? I'd like to pull this in the main branch and start testing with it. Thanks!

[kazuho]

My apologies for the huge delay. I have been busy on other stuff, but I'm finally getting back to working on improving QUIC throughput. I've taken a quick look at the PR and I think the design of the API is fine. Will merge this pretty soon. Separately, I like SEARCH and think that we should experiment with it. At Fastly, we have created (finally) the capability to run A/B testing using production traffic, which I am using now for evaluating Careful Resume. I will look into if we can use that for evaluating SEARCH.

[AmberCronin]

Hi, that's great news. Let us know if there is any way we can support the process further. Thank you!