Proposal and evaluation of packet loss rate inference scheme by using packet train method for active measurement

摘要

In recent years, the Internet has manifold traffic such as audio and video which require different QoS (Quality of Service). In the QoS aware Internet, performance measurement may become important technique to estimate network condition prior to data transmission and to confirm its quality. In the active measurement which is one of the measurement techniques, some observer sends probe packets, gets their performance such as packet loss, delay and/or throughput and infers some performance of the system without sending probe packets from them. Some amount of probe traffic is required to improve the accuracy of inference, however; this may disturb the status of the original system. Thus, in this paper, we will focus on the "packet train method" for sending probe packets, and new inference scheme in terms of packet loss rate by using the incremental behavior of loss rate of successive probe packets to moderate the inference of the disturbance with high precision. By modeling some router as simple Batch Geom/D/1 queueing system, we will theoretically and statistically analyze loss performance and investigate the effectiveness of the proposed scheme.