On 8/2/06, Andy Ruhl <acruhl@gmail.com> wrote:
> On 8/2/06, Andy Ruhl <acruhl@gmail.com> wrote:
>
> Ahh, as I read this I'm feeling like it's a troll post, and I wrote it!

In response to myself, just so this doesn't seem like a total troll,
here's one of the things I read yesterday after I googled TCP checksum
failure:

http://portal.acm.org/citation.cfm?id=347561&dl=ACM&coll=portal&CFID=15151515&CFTOKEN=6184618

One of the private responses I got pointed out that one of the authors
is a NetBSD developer, which I didn't know while I was reading it.
Very cool. It's only 10 pages. I don't understand the entire thing,
but I get the point of it. Here's the scary part, this is the last
sentence of the abstract:

"Even so, the highly non-random distribution of errors strongly
suggests some applications should employ application-level checksums
or equivalents."

"some applications". Probably meaning not a web browswer displaying
http to the screen, but probably everything else? I don't know...
Seems like any TCP app that's trusting the network falls into that
category in these days of multiple gigabyte and even petabyte backups.

Andy