Hi, In line:
> On Mar 31, 2017, at 9:59 AM, Neal Cardwell <[email protected]> wrote: > > On Fri, Mar 31, 2017 at 10:46 AM, Michael Welzl <[email protected]> wrote: >> Hi, >> >> As a researcher with possible interest in evaluating BBR, my question would >> be: what is “it” ? >> >> My understanding is that we have: >> - 2 IETF presentations >> - A paper in ACM Queue >> - Linux code >> >> and none of these three things are really equal. ( Sorry if I’m wrong! >> Maybe it’s only the ACM Queue paper that doesn’t quite match the rest? ) >> If this is true, it would be good to try to write the whole thing up in a >> draft so we could have a reference description of what BBR really is. >> >> Else, what would you consider the most important reference among the three >> items above? The Linux code, I guess? > > Yes, the Linux TCP code would be the most important reference, since > that's the most precise, and it's what's actually controlling YouTube > and google.com TCP. > > That said, the Linux TCP code should very closely match the CACM/Queue > paper. I can think of only one or two details in the gain cycling code > that were added to the code in the middle of the paper editing > process, and thus were not captured in the paper. And in turn, the > ICCRG presentations should also very closely match the code and paper > (except for the parts marked "experiments", eg the IETF 98 ICCRG > slides 14-19). Please let us know if there are specific places where > there are confusing divergences. That’s great to hear - my impression of differences was a mix of “impression from looking at things” and “impression from what people say” - so this confirmation that things don’t really diverge so much is very useful, thanks! > We are working on writing a draft or similar prose spec for those who > would prefer to use that as a reference. Wonderful ! Thanks for that too! Cheers, Michael
