End-to-End Loss Differentiation Algorithm Based on Estimation of Queue Usage in Multi-Hop Wireless Networks

Abstract

When TCP operates in multi-hop wireless networks, it suffers from severe performance degradation. This is because TCP reacts to wireless packet losses by unnecessarily decreasing its sending rate. Although previous loss differentiation algorithms (LDAs) can identify some of the packet losses due to wireless transmission errors as wireless losses, their accuracy is not high as much as we expect, and these schemes cannot avoid sacrificing the accuracy of congestion loss discrimination by misclassifying congestion losses as wireless losses. In this paper, we suggest a new end-to-end loss differentiation scheme which has high accuracy in both wireless loss discrimination and congestion loss discrimination. Our scheme estimates the rate of queue usage using information available to TCP. If the estimated queue usage is larger than 50% when a packet is lost, our scheme diagnoses the packet loss as congestion losses. Otherwise, it diagnoses the packet loss as wireless losses. Because the estimated queue usage is highly correlated to congestion, our scheme has an advantage to more exactly identify packet losses related to congestion and those unrelated to congestion. Through extensive simulations, we compare and evaluate our scheme with previous LDAs in terms of correlation, accuracy, and stability. And the results show that our scheme has the highest accuracy as well as its accuracy is more reliable than the other LDAs.