We have used both simulations and real world experiments to do the evaluation. Using the NS2 network simulator, we can examine UDT's performance in a broader set of network settings. However, we believe that real world experiments bear a higher significance in evaluating a network protocol. We tested our UDT implementation in many Gb/s and 10Gb/s network testbeds, especially the NCDM's teraflow testbed.
We tested UDT over a large variety of network situations, including different topologies, bandwidth, RTT, etc. We also changed the number of parallel flows.
Using the quantitative approach I just described, we can compare UDT's performance against other protocols, especially TCP.
In addition, we have used UDT in several real world applications. Finally, the UDT protocol has also been tested independently by many third party institutions. They have published some papers and technical reports describing their testing.