[ih] order of source and destination address in IP header

David P. Reed dpreed at reed.com
Fri Mar 25 06:45:59 PST 2005


I don't remember destination ever being first, nor do I remember anyone 
proposing the "performance benefit" that might have obtained.   That 
doesn't mean that such a consideration didn't motivate one of the 
intermediate references.

I'd like to amplify and clarify "state of mind" regarding TCP in the 
early days related to such matters.   The historians need to understand 
that there are many Johnny-come-latelies who seem to think that TCP and 
IP were invented as "network protocols" rather than protocols for a 
heterogeneous "network of networks" that each implemented their own 
low-level headers, ...   There was no notion of IP as the protocol that 
would be implemented at the switch level in the project at any time.

TCP and later the IP layer was expected to be a protocol *between 
gateways* at all the points in time being described.

I do know (because my predecessor protocol called DSP at MIT did it, and 
our work on source routing argued against it even having a clear 
meaning...) that we had some occasional discussions about why include a 
"source" at all in the header. Most protcols that might be used to 
implement IP routing didn't really use a source, and the source 
introduced another information channel that routers didn't really use.

Those of us in the security business viewed the creation of covert 
channels as worth tightening down (remember that part of the reason for 
splitting checksums and ports out of IP was our interest in being able 
to encrypt the payload end-to-end).

But in the end, including a source in a standard place in the IP header 
(rather than in the TCP header where it was needed for 
connection-establishment protocols to demux WKS session creation) was 
useful for practical engineering/debugging reasons.   Many of those who 
were going to have to debug the damn inter-gateway routing algorithms 
really needed some kind of tracing info that could be dumped about 
packets flowing through the network to figure out what was going on...





More information about the Internet-history mailing list