[IP] A Brittle Internet

[David Farber <dave@farber.net>]

Begin forwarded message:

From: "David P. Reed" <dpreed@reed.com>
Date: October 31, 2008 2:21:10 PM EDT
To: RJ Atkinson <rja@extremenetworks.com>
Cc: dave@farber.net, ip <ip@v2.listbox.com>
Subject: Re: [IP] A Brittle Internet

Ran - I find the same tendency to misunderstand packet communications  
at the wireless PHY layer, among many engineers.  To wit, in packet  
radio, you can't "measure the channel" when you are not transmitting,  
and the channel varies enormously during the time between successive  
packets from point A addressed to point B.

Yet, and this is maddening, there are a zillion CS-educated "wireless  
networking" types who think that it's a good idea to do "adaptive  
channel modeling" on a pairwise basis to "optimize" WiFi MACs.

Well, I have news:   adaptive techniques work on continuous RF  
transmissions operated in full duplex mode.  But it's silly on the  
face of it to presume that the channel between point A and point B is  
the same after A talks to C, to D, to E, and then talks to B again.   
Especially indoors.

Same for "dynamic power control" on a time scale of milliseconds to  
adjust rate.

Yet, these "circuit-based" ideas keep getting applied to packet  
systems radio phy's.   Is it because the professors who teach the CS  
networking courses are teaching radio as it was when they were in grad  
school 25 years ago and took the one EE course they've ever taken?

It turns out that what makes packet radios work is that the  
synchronization can be done on each individual packet, so the channel  
stability need only be presumed for a few milliseconds, and measured  
in less than a microsecond.

David Farber wrote:
> Begin forwarded message:
>
> From: RJ Atkinson <rja@extremenetworks.com>
> Date: October 31, 2008 10:05:34 AM EDT
> To: Dave Farber <dave@farber.net>
> Subject: Re: A Brittle Internet
>
> Dave,
>
> A curious phenomenon of the past decade or so has
> been the increasing brittleness of the Internet.
>
> Folks with a circuit networking background seem
> to be fundamentally uncomfortable reasoning about
> packet networking.  One result is the increasing demand
> for, and deployment of, circuit networking overlays,
> such as MPLS L2 VPNs, MPLS L3 VPNs, PBB-TE/PBT,
> and other things.[1]
>
> Use of these technologies greatly increases both
> the deployment and operational costs of the network,
> while visibly reducing the resilience of the network.
>
> Fifteen years ago, a fibre cut (e.g. Northridge earthquake)
> resulted in IP routing dynamically discovering new paths
> and a remarkable degree of Internet connectivity remaining.
> Today, a similar event would leave many folks disconnected,
> due to their reliance on circuit-like deployments over IP,
> even though an IP path remained, would be discovered by
> IP routing protocols, and could be used except for the
> reliance on circuit-emulation technologies of various
> sorts.
>
> It seems to me that somewhere the overall educational
> system has some gaps.  Communications techs/engineers
> coming out of technical school or college/university
> ought to have learned about packet networking and be
> comfortable reasoning about it.  Sadly, that seems
> often not to be the case at present.
>
> Yours,
>
> Ran
> rja@extremenetworks.com
>
> Disclaimer: Views expressed are the author's own;
> he never speaks for his employer.
>
> [1] Note that this use of MPLS is a bit different from
> "tactical" use of MPLS for short-term Traffic Engineering
> whilst waiting for new capacity to be brought online
> someplace.
>
>
>
>
>
>
> -------------------------------------------




-------------------------------------------