Bloat list:

I note that yesterday's problem with the long period latency swings was
narrowed down to a bad card or connector in the chassis, and that fiber
connection has been disabled.

I've got packet captures covering a goodly percentage of the universe now
(thanks to the hard work of the cc'd).... and occasionally I do things like
checking ipv6 vs ipv4 performance going internationally so as to have a
reference later.

http://gathering-edge.bufferbloat.net/~d/california/california_46compete-3.svg

Interesting about the ping part of the plot - there seems to be some
classification on the link, and it's neat
that BE traffic stops (packet loss there), that the EF traffic has got a
different RTT than the BK (CS1) traffic (which is the lowest latency of the
bunch!!!) , and that ipv6 has lower latency than ipv4 on this path(s). (duh)

as for the periodic bits of long period TCP changes, darned if I know...
that's what I have packet captures for.

you can see some sort of major packet loss event at T+168 or so here:

http://gathering-edge.bufferbloat.net/~d/california/california_46compete-long-3.svg

As to where it occurred, well, darn it,  "mtr" was part of the rrul spec
and haven't got around to integrating it.

Anyway, another long period 6vs4

http://gathering-edge.bufferbloat.net/~d/california/california_46compete-long-600.svg


-- 
Dave Täht

Fixing bufferbloat with cerowrt:
http://www.teklibre.com/cerowrt/subscribe.html