Re: [Cake] lanman2018 cake talk ideas

[Dave Taht <dave.taht@gmail.com>]

doing a self review of what I got before I crash

yes, yes, all the slides are out of order. Yes, I should write down
what I intend to say and not have it on the slide

* cake stores one less packet than htb+fq_codel. Anybody got a 2mbit
result for sqm that shows the resulting induced latency?
(2mbit because I'm irked about that 2mbit broadband study in the uk)
* I'll probably substitute my 50x1 ack-filtering result for the 30x1
in the paper (http://blog.cerowrt.org/post/ack_filtering/ ) and burn
no more than a minute on it because I still hate the idea of ack
filtering
* sch_cake (shaped 1gbit ethernet) v sch_fq bql'd at 1Gbit - for A
week NOW! I thought the bandwidth hit from shaping was 1.7 percent.
And I should know better! I realized about an hour ago that it was
(mostly) because we were getting 20% more measurement flows. I pulled
that data out of flent, but perhaps an irtt result shaped to a gbit
would be better (I didn't have irtt on that test). Also that test was
host to host, not through - which I kind of liked. It would probably
irk eric dumazet somewhat to see that result vs sch_fq :)
* I liked mentioning our most remote test site (mauritius), and wanted
to have something that talked about our development process near the
close
* need a final summary slide
* really hate the cake_dst/src/triple isolate graph and wish we had a
better way to express it. pie chart?
* think zooming into 300 flows with a probability of 0 collisions on
the 8 way set associative slide (and yes, the title needs to be
better)
* not sponsored by still motivates me every time it passes by
* liked explaining diffserv's implementation (bandwidth reservation
rather than drop probability)

I figure we have 25 slides tops