Connection Characteristics Analyses Based on Access Network Data

Abstract

This paper mainly considers the analytical mechanism for Internet connection characteristics. We extract the real packet level trace from an access network border router. No sampling method is adopted to maintain the full information. Parameters inside connections are examined carefully. Different time bins are detected for comprehensive exploration. Then, the potential relationship between IP addresses and port numbers at both sides are investigated two-by-two. Thirdly, in order to figure out the properties of connection size and duration, the fluctuation of packets volume is demonstrated, together with connections number. A new scheme based on dynamic window is proposed to calculate duration time efficiently. The result has shown several interesting phenomenon: 1 minute is a suitable bin for capturing the pattern in macroscopic perspective, less connections might generate more packets, etc.