44.2 Connection Vectors

A connection vector is a representation of an observed TCP connection. Each connection vector has several fields which specify the type of connection (sequential or concurrent), connection start time, loss rate, window size, etc. Each connection vector also contains an arbitrary number of application data units (ADUs). Each ADU represents data to be sent over the network by either the initiator or the acceptor of the TCP connection. ADUs also specify when data should be sent relative to either the reception of an ADU or the completed sending of a previous ADU. A file containing an arbitrary number of connection vectors must be supplied to Tmix in order for any traffic to be generated.

Connection vectors are either sequential or concurrent. Sequential connection vectors are the most common and represent a request-reply type connection, where a new ADU is sent only after the previous ADU has been received. Concurrent connection vectors allow for deviation from the sequential pattern. For a concurrent connection, new ADUs are sent a certain amount of time after the previous ADU was sent and do not depend upon when ADUs were received. Examples of this type of application protocol include HTTP/1.1 (pipelining) and the BitTorrent file-sharing protocol.

Tmix in ns-2 can handle two different connection vector formats, which we term as original and alternate. We have provided a Perl script (ns/tmix/cvec-orig2alt.pl) to convert from the original format to the alternate format. The Tmix module in ns-2 can automatically detect the format of the given connection vector file.