On Mon, 8 Dec 2025 17:46:43 +1100
David Gibson
On Mon, Dec 08, 2025 at 01:22:08AM +0100, Stefano Brivio wrote:
Patch 2/9 is the most relevant fix here, as we currently advertise a window that might be too big for what we can write to the socket, causing retransmissions right away and occasional high latency on short transfers to non-local peers.
Mostly as a consequence of fixing that, we now need several improvements and small fixes, including, most notably, an adaptive approach to pick the interval between checks for socket-side ACKs (patch 3/9), and several tricks to reliably trigger TCP buffer size auto-tuning as implemented by the Linux kernel (patches 5/9 and 7/9).
These changes make some existing issues more relevant, fixed by the other patches.
I've made a number of comments through the series. I think they do want some consideration. But given this works to address a real problem empirically, I'm fine for this to be merged as is, with more polishing later.
I just posted v3 addressing a few items that actually look wrong / problematic to me. Note that I forgot to change the "scaling" function name later in the series (as posted), I'm fixing it up on merge. -- Stefano