Lecture Notes for 10/14/99
Paper:   End-to-End Internet Packet Dynamics  [Pax97]

Prof. John Byers

Recap:
                   End-end measurement tools
                   Key principle:
                   [Bol93]  Bottleneck queues: space reserved all the way through  the propagation

                   [Pax96]  1.why network probe program
                                    2.pair measurement
                                    3.the difference of persistence/prevalence:
                                             persistence: how long will it last
                                             prevalence: default router
               (please see notes on 10/07/99 for details)

                 (* end-end dynamics: also Bolot's study *)

Interpacket space :    (irregular  but   userful)

Measurement: interarrival time: 2 hour
            .packet loss
            .packet delay
            .anomalies
What unexpected (wrong) behavior could one measure:
            .out of order 
            .routing loops: unobservable,except very long delays
            .duplicate packets(rare):arrive twice (reproduce in the network)
            .packet corruption:faulty-copy in buffer)
 

• out of order

       Why?          .routing changing  . other ways
           Impact of reordering:
                False retransmission:for example,packet 1 is very slow,so when packet 2,3,4 arrive (3 duplicate acks),
                it will trigger  retransmission(false!) ,since packet 1 will arrive finally ,instead of being dropped.
               Question: is 3 the correct times?

• corruption

• Bottleneck Bandwidth

• Packet loss

            unconditional loss: no assumption
            conditional loss:  assume    j-1 packet loss
 
            .Bolot: independent ,no correlation
           .Paxson:  consecutive losses (bursts)
                     Please see figure 8 in paper

• Packet delay