부산대학교 도서관

ISPs and enterprises usually overprovision their networks as a means of supporting QoS. In spite of that, the service quality of QoS-sensitive applications such as VoIP, video conferencing and streaming media may not be up to expectations. We believe this is due to sporadic but non-negligible packet losses due to traffic bursts even in underutilized links. Our earlier work attempted to detect and analyze packet losses in underutilized links in an enterprise network environment using SNMP. This paper presents an extension of our earlier work by attempting to detect and analyze packet loss at finer granularity of time scale. We have developed a passive traffic capturing system, which can provide smaller time scale analysis of packet loss. We have analyzed the data and identified parts that are representative of packet loss across various time scales: 10 milliseconds, one second, 10 seconds and one minute. Analysis reveals that packet losses on underutilized link do occur due to bursty traffic packets in a small time scale. We also present analysis of other traffic properties such as packet size distribution and flows for the packet loss.