🚀 go-pugleaf

In article <1990Jun18.220658.4369@druid.uucp> darcy@druid.UUCP (D'Arcy J.M. Cain) writes:
>I have been receiving duplicate postings recently but with one of them
>slightly garbled.  From examining the different path lines, it seems to
>be a problem at one of the following sites:
>[...]
>Anyone else seen this problem?  Here is an example message followed by the
>"bizzaro" version.  As you can see it seems to shift some letters by one
>in the alphabet and always in the same direction.  As you can see this
>happened in the message ID which is why it wasn't tossed along the way.
>It appears as two different messages.

A standard problem.  A compressed batch somewhere got a TINY noise hit...
when uncompressed, letters get shifted....  The noise was almost
certainly in uucp, as the error rates on main memory or disk are
typically MUCH lower.

As to how uucp missed the error, I have no idea, but a 16 bit CRC is
not God.  When 7MB of compressed news gets shipped through 10,000
machines (let's be conservative...but God not the BM kind!) you get
about a terabyte of data per fortnight, in about .5 trillion 64 byte
packets per year passing through uucp's connected to Usenet!

That's over a billion packets per day...with a packet error rate of
0.006% we would get one undetected error, somewhere on Usenet, per
day.  I originally wrote:
	The rate's probably better than 0.006%, but it's probably worse
	than 0.00002% so we'll get at least one such error, somewhere, per year.
but on second thought, I doubt the error rate IS much better than that
(one in 16,000).

As for where it actually happened, it's pretty hard to say.  Most of a
batch (as it existed SOMEWHERE along the path to you) will suffer the
error, and these articles stand a fair chance of having similar paths,
which can lead to erroneous conclusions.  Also, because they have
mangled Message-ID's they will make it all the way through the net.  I
too have seen a fair number of articles with the problem.  It would
NOT have happened over an NNTP link, though, as that would only affect
a single message, as NNTP doesn't (as far as I know) batch or
compress. 

For the record, the article you mention was also mangled here.  The
first similar point in the path is torsqnt, so all you can rule out
for sure is the connections from torsqnt to you, plus the NNTP links:
	!torsqnt!			- could have happened here
	news-server.csri.toronto.edu!	- NNTP link
	clyde.concordia.ca!		- NNTP link
	uunet!				- NNTP link
	cs.utexas.edu!			- could have happened here or earlier
					  (I presume this isn't NNTP)
	usc!apple!voder!pyramid!decwrl!ucbvax!van-bc!sl

And I bet THAT was more answer than anybody wanted.... :-)

	../Dave