To disambiguate the multiple possible interpretations of the t value with sequential connections, we have modified the connection vector format into an alternate format. The basic idea is that in the case of sequential connection vectors, there really exist two times associated with the initiator or acceptor while sending an ADU to the other side:
This same alternate representation can be used for concurrent connection vectors also. But because there is no time dependence between the sides of the connection, each side schedules sending its next ADU with respect to time at which it sent its last ADU. Therefore, in case of concurrent connection vectors, recv_wait_time is not applicable and is always set to 0. Also, in case of concurrent connection vectors, both the sides start sending their messages at the same time.
We keep the header lines of the connection vectors (those containing the start time, window size, RTT, and loss rates) in the same format as the original, except that we replace SEQ with S and CONC with C. Lines beginning with I denote actions for the initiator, and lines beginning with A show actions for the acceptor. The remaining format of these lines is send_wait_time recv_wait_time bytes. Figure 44.3 shows the alternate connection vector format corresponding to the sequential and concurrent connections shown in Figure 44.2.
Tom Henderson 2011-11-05