A few days ago I wrote an item entitled How Soon We Forget (Technology).
In the interest of reviving lost ideas I've obtained permission from Cisco to revive some work that I had left unfinished and unpublished. My intention is to refine and extend this idea and submit it to the IETF.
I was, and continue to be, interested in the issue of network control. For example see my talk From Barnstorming to Boeing - Transforming the Internet Into a Lifeline Utility (powerpoint) (speaker's notes in .pdf)
One piece of this continent-sized chunk of internet terra incognita are mechanisms to learn the shape and quality of the pathways through the net. This information is necessary for troubleshooting, content management and placement, and service level assurances.
It is possible to squeeze a lot of useful information out of tools like Traceroute and pchar, and through inspection of internal and external (BGP) routing information. (For example, take a look at CAIDA's Macroscopic Topology AS Adjacencies. But more precise data and more extensive information is needed.
I was intrigued by Pathcar and its sucessor pchar. But these tools depend on measuring the external behaviour of routers; these tools deeply refine and analyze the noisy raw data and winnow out the information. Having come from the SNMP community I wondered whether it would be easier, faster, more efficient, and more accurate to simply see if the information could be obtained directly from the routers along a pathway. The mechanisms used in mtrace struck me as a useful architectural approach.
The result is FPCP - the Fast Path Characterization Protocol.
I stopped work on it back in year 2000 because running-for and being-on the ICANN board of directors began to consume too much of my time.
My stopping point from several years ago is visible at the following URL: /archive/fpcp/fpcp-sept-19-2000.html
I am going to pick up where I left off and turn the existing draft into a more complete proposal that can be sent to the IETF. If anyone is interested in helping in this work, please let me know.
Posted by karl at January 20, 2005 2:31 PM