From the combined traffic type dependent tests at the Force10 Loop-Back, the following conclusions can be drawn:

• In nearly all TCP tests that were executed as a function of the TCP window size, the # source / destinations and the # parallel flows between two node pairs the provided bandwidth of the line (bottleneck: 1 Gbits/s at the loop-back) could be achieved.
• Also the UDP flows could well utilize the bandwidth capacity. However, especially with multiple parallel flows, some 10% - 20% of the connections got broken. The cause for that connection loss is not yet clear.
• In the combined UDT and TCP tests the UDT flow is adjusting itself almost complete to the multiple TCP flows.
• When combined with TCP flows, the RBUDP flow is only partly adjusting itself to the TCP flows, but there never has been claimed that RBUDP should be TCP friendly.
• In the case that an UDT or RBUDP flow has been combined with multiple TCP flows there is already adjustment of the special protocol flow to the Iperf flows before the provided bandwidth limit has been reached.