If we're starting the discussion on Multipath, it's likely that starting
new email threads would be a good idea.

Beyond that ...

On Wed, Sep 30, 2020 at 5:28 AM Olivier Bonaventure <
[email protected]> wrote:

> The main benefit of putting multipath inside the transport is that the
> application does not need to handle the problems that occur on any of
> these paths. The transport has access to round-trip-times and loss
> measurements and has all the information required to manage the
> different paths efficiently.
>

>From the beginning of time, people who needed some, but not all, of TCP's
transport capabilities were left using UDP, and figuring out the transport
capabilities in their own applications, with greater or less great success.

The quote I've heard so many times is that many UDP-based applications
ended up adding so many TCP-like transport capabilities that they might
just as well have used TCP.

We've had to write BCPs about that - guidelines for UDP usage, with
https://tools.ietf.org/html/rfc8085 replacing
https://tools.ietf.org/html/rfc5405 ten years later being the most obvious
example, but also circuit breakers, like
https://datatracker.ietf.org/doc/rfc8083/ (a Proposed Standard) and
https://datatracker.ietf.org/doc/rfc8084/.

Meanwhile, we've had to try to solve the "some but not all of TCP" problem
in transport ANYWAY - first with SCTP and DCCP, and now with QUIC.

So, I was hoping that we could learn from that example, and do work on
multipath within transport, so that the world would not be full of
applications managing multiple paths with greater or less success.

Do the right thing, of course.

Best,

Spencer

Reply via email to