In the attached 5 minute plot from a few days ago (I can supply the
flent.gz files if anyone wants them), I see a puzzling spike at T+155s
to nearly 90ms of baseline latency, then down to 20ms. No degree of
orbital mechanics can apply to this change, even factoring in an over
the horizon connection, routing packets on the ground through LA to
seattle, and back, or using a couple ISLs, can make this add up for
me. A combination of all that, kind of does make sense.

The trace otherwise shows the sawtooth pattern of a single tcp flow ,
a loss (sometimes catastrophic) at every downward bandwidth change.

An assumption I have long been making is the latency staircase effect
(see T+170) forward is achieving the best encoding rate at the
distance then seen, the distance growing and the encoding rate falling
in distinct steps, with a fixed amount of buffering, until finally
that sat starts falling out of range, and it choses another at T+240s.

But jeeze, a 70ms baseline latency swing? What gives?  I imagine
somehow correlating this with a mpls enabled traceroute might begin to
make some sense of it, correlated by orbital positions....


-- 
Oct 30: https://netdevconf.info/0x17/news/the-maestro-and-the-music-bof.html
Dave Täht CSO, LibreQos