<<  Protocol Testbed Results   |   Table of Contents   |   Standard TCP / UDP Tests  >   |   Force10 - 6509 - AR5 Loop  >>

Force10 Loop-Back: The Zero RTT

Setup

As zero round-trip time test a local loop-back has been configured at the Force10 switch with a provided bandwidth of 1 Gbits/s. The DAS-2 nodes, all located in the same subnet, were separated in two VLAN's that were connected via the local loop-back. This topology is displayed in . The small provided bandwidth of 1 Gbits/s makes this topology suited to test the congestion avoidance behaviour of the alternative transport protocols when combined with standard TCP protocol flows.

 +-------+
 | DAS-2 |-\   VLAN A
 +-------+  \
   .....     \   +---+
 +-------+    \--|   |
 | DAS-2 |--\    | F |
 +-------+   \---| O |
                 | R |---\
- - - - - - - - -| C |-  |  Loop-Back
                 | E |---/
 +-------+   /---| 1 |
 | DAS-2 |--/    | 0 |
 +-------+    /--|   |
   .....     /   +---+
 +-------+  /
 | DAS-2 |-/   VLAN B
 +-------+

.    Topology of the used connection scheme. The DAS-2 nodes, all located in the same subnet, were separated into two VLAN's which are inter-connected via a loop-back.

Performed Experiments

Between the both DAS-2 VLAN's networks tests with multiple flows were executed. First the behaviour of multiple standard TCP and UDP flows were investigated. As always, unless stated elsewhere, the Iperf traffic generator V. 1.6.5 is used to create the TCP and UDP flows. Hereafter, the influence of the alternative transport protocols upon small TCP flows is investigated, the so called TCP friendliness. These experiments are generally executed by starting one large alternative protocol flow and after some time a large number of small TCP flows, or the other way around. To generate TCP flows with the largest possible diversity, each participating node in one VLAN will send in general flows to all participating nodes in the other VLAN.

To be able to generate small TCP flows, also in the situation with zero RTT, the Iperf tool has been modified such that also a TCP shaping bandwidth can be specified, comparable with the shaping option for UDP. Due to the bursty character of TCP there is no guarantee that also in a small timescale a flow is not bursty, but by using a large number of streams we expect some averaging of the burstiness.

In the next sections the results of the various experiments will be presented.


<<  Protocol Testbed Results   |   Table of Contents   |   Standard TCP / UDP Tests  >   |   Force10 - 6509 - AR5 Loop  >>