next up previous
Next: Basic Bottleneck Configuration Up: Improving TCP Congestion Control Previous: Proposed Implementation

   
Performance Results

In this section we examine the performance of TCP-SC compared to TCP Vegas [3] and TCP Reno [19]. We first show performance results for a basic configuration with a single source and a bottleneck link, then a single source with cross-traffic on the reverse path, and finally performance over asymmetric links.

We have measured performance for TCP-SC through simulations using the ``ns'' network simulator [16]. The simulator contains implementations of TCP Reno and TCP Vegas. TCP-SC was implemented by modifying the existing TCP-Reno source code to include the new congestion avoidance and error-recovery schemes. Unless stated otherwise, data packets are of size 1Kbyte, the maximum window size, $cwnd\_max$ for every TCP connection is 64 packets (which corresponds to a receiver advertised window of 64K), and the initial ssthresh is equal to $\frac{1}{2}*cwnd\_max$. All simulations are an FTP transfer with a source that always has data to send; simulations are run for 10 seconds.



 

Chris Parsa
2000-01-25