The original DSR paper described datagrams as the most primitive signal thru the net. It provides no priority information however. I can imagine that some switches would offer some sort of priority on outgoing links as indicated by a “quality mark” within the routing information. In fact each switch determines the meaning of its portion of the routing field. It would be well to establish a standard here but not logically necessary. It is hard to imaging a standard on meanings of quality marks. The toll extracted by a switch could well depend on the quality mark. (But see ‘Bang-Bang’ below.)

It would be necessary for guides to deal with the meaning of quality marks, perhaps as empirically determined for each switch as part of their duty to advise travelers on the quality of the switches.

I think that most QoS issues are better handled for circuits rather than datagrams.

Just now (2012 June 12) it occurs to me that the portion of the steering data interpreted by a particular node is where you can put priority data. I had thought that the size of that information should be an early design freeze, perhaps 4 bits.

(2012 Sept 12) Priority packets might cost more. Dynamic pricing is a complex issue and complex derivative markets are possible. Certainly is should be possible to reserve transmission capacity. I do not propose that such software should reside in the switching nodes. Showing that you are a claimant to such a reservation is a curious problem. Here is some speculation. When I imagined that circuits were a primitive of the network they were easier to design.

(2012 Oct 3) Bang-Bang Prices I think the following convention solves many problems. A node registers two interface numbers for the same physical link with scouts. Packets routed over that link can specify either turn op. Two prices are advertised by the node for that link. I imagine a factor of 8 in price. The node always charges the lower price except in case of link congestion. Packets via the first turn op are reflected upon congestion. Packets via the second turn op pay the higher toll. Note the delicious temptation of the node to declare congestion. My intuition is that the values that network users put on packets will vary over orders of magnitude and that a good network will provide service at prices set by real competition, i.e. commodity prices. One might call this the bang-bang price strategy in analogy to the simplest servo designs.

(2012 Oct 23) Using turn ops to specify priority also solves the problem of packets delivered in order. If you want a packet to overtake others that you have already sent, and while building your circuit you choose less than top priority, (likely for bulk data), then you can route a packet using the high priority turn ops which you probably learned of from the guide when the circuit was built. QoS always depends on the path and most software that deals with QoS needs to understand that, and perhaps it needs no concept of node identity.