[Discuss] Off-Topic [IP] BufferBloat: What's Wrong with the Internet?

[richb at pioneer.ci.net (Rich Braun)]

Bill Horne <bill at horne.net>
> At some point, the Internet will need a major overhaul.

Will it?

> For what common carriers are trying to do...TCP/IP can't be made to fit.

In 2001 I was working at a router company with a solution in search of this
problem.  The company blew through its VC funding while customers just threw
fatter pipes (and big iron from Juniper) at the problem.

Over time, I've become pretty well convinced that the latency problem can and
will ultimately be solved by fatter pipes, not fancy new protocols.  It's just
plain cheaper to put monitoring in place on a link, and route around it
between upgrades, than to overhaul TCP/IP.  And even with the occasional
glitches, the system works vastly better than the phone system ever did in its
pre-ESS days.  And it got that way over a 20-year period, not 120 years.

> This fight will be
> about which mega-corporations carve out virtual slices of Internet
> bandwidth so that they can avoid paying for their own.

That part I agree with.  But I think those mega-corps will be running IPv4 on
their backbone links for the next 10-15 years, even if China gets going with
IPv6 in the next 3-5 years.  Few of them are holding board-level discussions
about the claimed "BufferBloat" problem. ;-)

The gearheads who recognize BufferBloat will ultimately do the obvious: crank
down the buffering and adjust the retry parameters.  And flatten out the
number of hops from source to destination.  I worked at a disk-drive company
back in the 1990s that had to deal with virtually the same issue:  on a
contract with Time-Warner to ship disks for video-on-demand disk arrays (back
when I/O performance for video was a bottleneck), we had to adjust and/or gut
out the retry logic to punt on bad sectors rather than impact latency.

-rich