Corrections to How Does TCP Generate Pseudo-Self-Similarity? appeared in MASCOTS 2001 Liang Guo guol@cs.bu.edu Mark Crovella crovella@cs.bu.edu Computer Science Dept. Boston University Boston, MA 02215 Ibrahim Matta matta@cs.bu.edu In this note we would like to clarify and amend a number of points made in [3]. One of the central points in [3] was that although TCP s exponential backoff algorithm can introduce noticeable correlations in traf c from individual ows, the resulting correlation structure is not selfsimilarity. The term pseudo-self-similarity was used to describe this correlation structure. This term, introduced in [4], was rst used in [1] to describe the same correlation structure in TCP ows. An early version of our paper appeared as [2]. In [2], although we pointed out from simulations that the limited retransmission timeout (RTO) in TCP s exponential backoff causes effects only over limited timescales, we did not make clear that TCP s correlation structure was not self-similarity. Adding this point was the principal improvement in [3] over [2]. In [3] we wrote that [1] extended our model which does not credit [1] in two ways. First, [1] was the rst to correctly and forcefully make the point that the correlation structure is not self-similarity due to limited RTO in exponential backoff. Second, the Markovian model in [1] was independently derived, is more detailed, and can be used to illustrate the correlation structure of both exponential back-off and congestion avoidance phases of TCP. We regret these inaccuracies in our paper. Pseudo-Self-Similarity? In Proceedings of the International Workshop on Modeling, Analysis and Simulation of Computer and Telecommunications Systems - MASCOTS 01, Cincinnati, Ohio, August 2001. [4] S. Robert and J.-Y. L. Boudec. On a Markov Modulated Chain with Pseudo-Long Range Dependences. Performance Evaluation, 27&28:159 173, 1996. 1.
/lp/association-for-computing-machinery/corrections-to-how-does-tcp-generate-pseudo-self-similarity-KuFkVYZWg1