[e2e] It's all Farber's fault

David P. Reed dpreed at reed.com
Tue May 15 07:04:37 PDT 2007

Lloyd Wood wrote:
> Actually, source routing comes out of the confluence of identity and location in address. If we had clear separation there, we wouldn't have source routing...
That's ridiculous.   All of the arguments for source routing (or almost 
all) relate to the idea that its a good idea to move function out of the 
core of the net to enable flexibility.  That is especially true of 
source routing.   ARPA had requirements for catenets that involved *not 
routing certain traffic through certain subnetworks*, for example.  To 
achieve that  by creating 100's of millions of  "classes of service" 
with different routing tables in the  network for each class of service 
would have been impractical.

Source routing is there as an option because (coupled with "control 
plane" or "knowledge plane" layer routing service) one can move the 
specialized routing needed by some traffic out of the network.

At one point in the history of the Internet (when I joined the original 
team) there were strong arguments for making source routing the 
*primary* form of network routing.   The arguments for it were captured 
in Jerry's and my note.   The word "campus" was included in the title 
largely for a sort of "political" reason - the campus internetwork was a 
greenfield interop space concept (LANs were really new, and the idea of 
computer-computer messaging, rather than remote login, were biggest in 
campuses like MIT and Xerox PARC - you had to be there to understand the 
cultural distance between the LAN guys with 1500-byte packets like me, 
John Shoch, and Ken Pogran and the character-at-a-time remote login guys 
like Larry Roberts and the BBN-TIP guys - that's why it was so hard to 
split out functions that should have been part of the TELNET protocol 
from TCP, and functions that should have been part of a reliable stream 
from the basic datagram transport).

More information about the end2end-interest mailing list