On 6/11/20 11:26 AM, Craig Partridge wrote:
Two comments on the several notes that have gone past.
Michael Thomas raises a point about bugs. He's quite right and
there's reason to believe that if we tracked the Internet paths on
which checksum errors took place, and compared the paths to find
commonalities, we'd locate broken/buggy equipment. Jonathan Stone did
some of that 20 years ago and found a software bug in a vendor TCP
(which the vendor updated in a matter of days -- kudos to them) and a
network interface vendor who was shipping cards with defective
hardware (and adamantly claimed our measurements identifying their
cards were wrong).
Ha! I once got sent on site to find a tcp bug for my stack which we
couldn't reproduce. Found it in about an hour or two. Got to see the 747
that transported the space shuttle. Now *that's* a transport layer you
don't want to find bugs in :)
There was also a note (which I can't seem to find) to the effect of
"the errors in the papers don't match what I'm seeing on my systems."
That's not a surprise. Errors cluster around particular links or
devices. There's also some reason to believe error rates are higher
on the fast links. So, depending on your environment and where your
traffic goes, your error rates may vary widely. All the more reason
to measure.
It also wouldn't be surprising that it highly correlates with version
1.0 vs 1.1. One of the depressing things about mobile phone stuff is the
degree that vendors stop supporting updates, with it being particularly
bad for Android.
Mike
