Peace,

On Thu, Jul 15, 2021, 4:40 PM Nick Banks <[email protected]> wrote:

> You should definitely not make any assumptions around having unique source
> ports for QUIC connections.
>

Ephemeral port randomization-based traffic balancing is not what I assume,
it's what, maybe, half the Internet connectivity providers do for a
decade.  RFC 6438 kinda sums it up, and I urge you to read it.

To distribute the load across multiple paths, an Internet router needs an
entropy source providing a random integer stable for a connection (if you
want to avoid delays and retries), generally different for different
connections (if you want to avoid seeing all the connections congested on a
single paths), available for reading in every packet.  Now, that's the
ephemeral source port.

Compared to a protocol X which ditches the ephemeral port, TCP would
probably work faster and more reliable along a lot of the network paths,
because it won't be always taking paths already congested by the protocol X.

That, of course, assumes the protocol X is popular. Many VPN protocols also
suffer from that but their general (negligible) Internet-scale adoption
state kinda saves the day.  For the HTTP protocol that would be different.

This doesn't necessarily mean all connections would be on the same source
> port, but only a few ports might be used for all connections.
>

*That* *might* work but the fixed source port won't.

--
Töma

>

Reply via email to