[e2e] Survivability

David P. Reed dpreed at reed.com
Mon Jul 28 12:47:11 PDT 2003


At 02:54 PM 7/28/2003 -0300, Fernando Gont wrote:
>If the Internet protocol suite was developed by the DARPA, with military 
>objetives, why was some survivability sacrified in order to be able to 
>interconnect *existing* networks?  I mean, why didn't the "single 
>multimedia network design" idea succeed?

Because interoperation is more useful than survivability of an "island" 
network that connects to nothing.   The military problem of interoperation 
is much bigger than perfect survivability of the network unto 
itself.   Interoperation makes the whole military survivable.


>If somehow it was thought that the Internet would be used, at some point, 
>for commercial activity, why was "accountability" (for example) *last* in 
>the list of goals?

Accountability (in the sense of allocating economic costs) is, in the end, 
about post hoc storytelling or blame-finding, etc.   The value of a network 
cannot be reduced to the cost of its components, and the value of each 
component is dependent on contingent future demand.   Loosely translated, a 
network creates a set of economic *options* whose value is contingent on a 
highly variable future set of conditions.   Those options cannot properly 
value the network and assign costs based on usage or whatever other measure 
without an assumed PDF over all future network demand.  In particular, 
options that do not get exercised (messages not sent) create value, just as 
options that do get exercised, because the point of the network is to 
enable communications among any pair of nodes that decide to communicate, 
rather than to enable communications only among those that are known to 
need to communicate at the time the network is built.
Allocating costs according to any particular scheme is a political act, not 
an engineering problem, and thus is best considered in a political science 
class, not an engineering project.

Put simply:   If you set out to design an accountable network, it's most 
likely you will design a network that accounts very well for nothing good. 




More information about the end2end-interest mailing list