[e2e] random early time-to-die

[Sebastian Zimmermann]

Murat Yuksel wrote:

> 
>>This is a form of congestion pricing using the TTL as a purse?
>>                                                                         -JMS
> 
> I think it is not "pricing". You are decreasing TTL for the purpose of
> defining when to kill the packet. Congestion based pricing would be to
> increase the price of that packet rather than killing it.

View it the other way around:
Each packet has a "budget". It doesn't matter whether it is encoded in 
the TTL field or in another yet-to-be field. Each router decreases the 
budget by an amount proportional to the "congestion charge". When the 
budget is zero, the packet is dropped.

But what can we gain from that? One of the ideas behind congestion 
pricing strategies is that sources adjust their rates without packet 
loss. This scheme still uses packet loss. So the known congestion 
pricing strategies seem to be superior to this proposal. Dropping the 
packets might provide an incentive, but then the sender should not be 
able to freely set its budget.

However, could we use this as admission control? Let's assume, the total 
cost for end-to-end delivery is roughly constant for short time periods. 
If the budget is lower than the cost, all packets are dropped = the 
connection is refused. If the budget is greater than the cost, all 
packets are delivered = the connection is accepted.
By setting a budget field in the packet header, a sender could choose 
the importance of the connection. Of course, the budget should somehow 
correspond to real money, otherwise everybody would choose the maximum 
amount. But one could implement service classes that way.

The questions remains, however, if we want admission control at this layer.


- Sebastian