[Iccrg] A "Thought Experiment"
john at jlc.net
Thu Apr 17 20:12:01 BST 2008
I would like to propose a thought experiment.
Let us assume an "ideal" internet where a large number of routers,
managed by a largish number of Autonomous Systems are interconnected
pairwise with bidirectional connections such that there is a data path
with a known minimum latency and maximum bandwidth, separate from a
control path of similar latency and infinite bandwidth; and let us
assume each router has infinite memory and infinite processing power.
(Obviously I don't propose to build such a network.)
This would, IMHO, be a good basis to start designing an "ideal"
routing architecture. For one example:
1) assume all routers have a UTC clock synchronized to a microsecond.
2) assume all packets are injected with a UTC expiration time.
3) assume routers will drop packets that "cannot" be delivered by then.
4) assume that drop event can be reported back over the control path.
We now have a much more useful measure of "congestion," in that the
sender can know, quite possibly before the packet would have been
delivered, that there's too much "congestion" for timely delivery.
S/he can then try to live with slower delivery or give up trying to
use this internet for those packets.
A number of optimizations present themselves for consideration,
a) higher priority could be assigned to packets which will expire soon.
b) projected latency (not just minimum) could be reported back the path.
c) an ISP could (per contract) switch packets to a more expensive path.
(We're even getting close to being able to route around congestion
_within_ the network.)
Does this sound interesting? If so, we might actually think about
building something with "relaxed" values for "infinite"... ;^)
John Leslie <john at jlc.net>
More information about the Iccrg