Bon Jour,
But as I am German, I will use English from now on ;-)
OK, there we go, I try to keep it as simple as possible:
A1 - WFQ orders packets based on "when they would be finished to send". CDT does not mean the maximum number of packets a queue can have. If the whole queueing system has more packets than CDT, packets who would be the last to send is discarded, if a new packet arrives. An example, for simplicity assume two queues Q1 and Q2 and two flows, one in each queue. Assume we have CDT of 6 and queue size of 7.
Assume we enqueue 6 packets of 100 bytes each in Q1. now a packet with 1500 bytes arrives for Q2 - and we will enqueue it! Why? WFQ will always enqueue a packet into an empty queue. Now we have Q1 6x100 bytes and Q2 1x1500 bytes.
If another 100 bytes packet arrives for Q1 WFQ will drop the packet from Q2 and we get 7x100 bytes in Q1.
At this point, if the queue size would have been 6 like CDT we would have tail dropped the 7th packet, which was entitled to be sent!
The scenario above is of course a bit oversimplified, as enqueueing and dequeueing are going on at the same time and we have much more logical queues, but I hope you can accept, that CDT is not the maximum number of packets in a single queue.
A2 - The wording sounds strange to me. IP Precedence is used with DiffServ and usually not with IntServ. However, I would hesitate to call it a QoS algorithm.
A3 - PHB is a term defined in the RFCs for DiffServ. The idea of DiffServ is scalability, thus no signalling (IntServ uses RSVP for signalling of QoS treatment), but marking of packets. As many different packets from different flows will be treated the same on a single hop, one can call it "Aggregation" and the QoS treatment of packets in a certain class is called "Behaviour Aggregate".
As there is no signalling each hop treats traffic independently of other hops. To get an end-to-end treatment of traffic, some "Per-Hop Behaviours" have been predefined in the RFCs. An example would be EF, which was defined with VoIP traffic in mind: minimize delay and jitter, rather drop than delay a packet, as "late" means "useless" in real time applications.
Cisco implementation: priority queue in LLQ.
A4 - If you use only 3 bits the rest is ignored. DSCP however uses the first 6 Bits of the byte formerly known as TOS and the last two can be used for ECN according to RFC 3168. The nice thing about DiffServ is the scalability and flexibility. IF you want, you can use various header field for classification of traffic. But you do not have to and can choose the level of complexity fittig your needs.
Hope this helps! Please use the rating system.
Regards,
Martin
