The issue raised by Mirja is here(though it was subsumed by an existing issue): https://github.com/quicwg/base-drafts/issues/4050
The quote of interest is "in https://tools.ietf.org/html/rfc5681#section-4.3 ssthresh is reduced upon entering recovery but cwnd is set to ssthresh when leaving recovery."
QUIC's old behavior was overly conservative, since it reduced the CWND immediately upon loss, rather than reducing ssthresh immediately and CWND upon exiting recovery. The new text allows for the old behavior or RFC5681 or things in between, with PRR being an example.
Ian
On Sun, Oct 25, 2020 at 8:42 AM Yoshifumi Nishida <nsd.ietf@xxxxxxxxx> wrote:
Hi Gorry,On Sat, Oct 24, 2020 at 4:36 AM Gorry Fairhurst <gorry@xxxxxxxxxxxxxx> wrote:
So I reviewed this ID several times in the WG. At the WGLC, I had no
substantive issues, and I see quite a lot of change since WGLC (most
really good!).
However, I also something I hadn't expected and that since rev -30, this
has introduced the option to use PRR:
..."Implementations MAY reduce the congestion window immediately upon
entering a recovery period or use other mechanisms, such as
Proportional Rate Reduction ([PRR]), to reduce the congestion
window
more gradually."
This did surprise me, but perhaps the working group thinks this is Reno
behaviour?This seems to be ok to me as QUIC's CC already deviates from Reno in various points and the doc just says "similar to Reno".Another point is this part is not mandatory.Since PRR doesn't change the window size after recovery, the deviation looks minor from my point of view.As long as the algorithm reduces the congestion window properly (be close to ssthresh at the end of recovery), I think it won't be aggressive.Thanks,--Yoshi
-- last-call mailing list last-call@xxxxxxxx https://www.ietf.org/mailman/listinfo/last-call
