Error Modeling for TCP Simulations

This work focuses on the implications for TCP performance with respect to two broad characteristics of error modeling. First, we consider rate-based error models vs. temporal error models. We show that it is difficult to confidently assess the impact of improvements to existing protocols or the performance gains of new protocols under a ratebased model. Second, we consider error models that “synchronize” losses in the forward and reverse paths vs. models that do not do so, or which model losses in the forward path only. Our results show that TCP performance can be dramatically different under a synchronized error model compared to a non-synchronized one.