Yep these are BBRv1 nodes with an older kernel (without even wifi patches[1]) so their TCP performance is quite suboptimal.
Once we update Edge to 5.4 we'll start the global BBRv2 rollout to all PoPs. In the mean time, when you hit www.dropbox.com you should expect (with a very high probability) that you are hitting a BBRv1 endpoint. [1] https://lore.kernel.org/netdev/[email protected]/#t -- t: @SaveTheRbtz > On Apr 25, 2020, at 8:10 PM, Neal Cardwell via Bloat > <[email protected]> wrote: > > > From: Neal Cardwell <[email protected]> > Subject: Re: [Bloat] dropbox, bbr and ecn packet capture > Date: April 25, 2020 at 8:10:11 PM PDT > To: Jonathan Morton <[email protected]> > Cc: Dave Taht <[email protected]>, ECN-Sane > <[email protected]>, bloat <[email protected]> > > > > > On Sat, Apr 25, 2020 at 8:52 PM Jonathan Morton <[email protected]> wrote: > > On 26 Apr, 2020, at 3:36 am, Dave Taht <[email protected]> wrote: > > > > I just did a rather large dropbox download. They are well known to be > > using bbr and experimenting with bbrv2. So I fired off a capture > > during a big dropbox download... > > > > It negotiated ecn, my fq_codel shaper and/or my newly ath10k > > fq_codel's wifi exerted CE, osx sent back ecn-echo, and the rtt > > results were lovely. However, there is possibly not a causal > > relationship here, and if anyone is bored and wants to scetrace, > > tcptrace or otherwise tear this cap apart, go for it. > > Well, the CE response at their end is definitely not Multiplicative Decrease. > I haven't dug into it more deeply than that. But they're also not running > AccECN, nor are they "proactively" sending CWR to get a "more accurate" CE > feedback. I suspect they're running BBRv1 in this one. > > Agreed, that looks like BBRv1. I see a few BBRv1 PROBE_RTT phases in there. > > Thanks for the trace! > > best, > neal > > > > _______________________________________________ > Bloat mailing list > [email protected] > https://lists.bufferbloat.net/listinfo/bloat _______________________________________________ Bloat mailing list [email protected] https://lists.bufferbloat.net/listinfo/bloat
