| ^ | Table of Contents | | | Force10 Loop-Back: The Zero RTT | > | | | Force10 - 6509 - AR5 Loop | >> |
In this document the results of various alternative transport protocols are compared as a function of the round-trip time. This will in general be done by measuring the impact from a fat, alternative transport protocol flow upon many small, standard TCP flows which are generated with a modified version of the Iperf traffic generator in the sense that this version is also able to shape the TCP bandwidth sufficiently. In general these many TCP flows are generated with the DAS-2 computer cluster from the Faculty of Science at the UvA. For this purpose the cluster nodes have been equipped with Gigabit Ethernet cards that are inter-connected via a Force10 switch.
The round-trip time will be varied by defining various routes between the DAS-2 cluster nodes by means of reconfiguration of the network equipment. In general the routes are defined from one half of the cluster to the other cluster half. For this purpose both cluster halves are logically separated by placing both halves in two separate VLAN's. Besides the round-trip time, also the TCP window size and the shaped bandwidth will be varied for the TCP streams. For the alternative transport protocols these parameters are parameter dependent, but sometimes they will include an equivalent of the TCP window size and often the block size that will be used to transfer the data.
Most analysed alternative transport protocols also supply besides some simple traffic generator tools also an API. This possibility has been used to implement for these tools traffic generators with an Iperf like look and feel concerning the program options and output. The major differences with Iperf are:
The following transport protocols will be treated in this document (see also this overview):
| ^ | Table of Contents | | | Force10 Loop-Back: The Zero RTT | > | | | Force10 - 6509 - AR5 Loop | >> |