This paper evaluates the accuracy of Poisson processes in modeling wide-area network traffic, showing that they often fail to capture the burstiness and self-similarity of real traffic. The authors analyze 21 wide-area TCP traces, examining various traffic patterns, including TELNET, FTP, and SMTP connections. They find that user-initiated TCP sessions, such as TELNET and FTP control connections, can be well-modeled as Poisson processes with fixed hourly rates. However, other connection arrivals, such as FTPDATA connections, exhibit significant burstiness and do not follow Poisson distributions. TELNET packet interarrivals, when modeled using exponential distributions, severely underestimate burstiness, while empirical Tcplib distributions better reflect actual traffic patterns. FTPDATA connections within FTP sessions are clustered into "bursts," with the largest bursts dominating FTPDATA traffic. The authors also find that wide-area traffic exhibits self-similarity, with burstiness persisting over many time scales. This has implications for congestion control and traffic performance. The paper concludes that self-similar models may better capture the characteristics of wide-area traffic than Poisson processes.This paper evaluates the accuracy of Poisson processes in modeling wide-area network traffic, showing that they often fail to capture the burstiness and self-similarity of real traffic. The authors analyze 21 wide-area TCP traces, examining various traffic patterns, including TELNET, FTP, and SMTP connections. They find that user-initiated TCP sessions, such as TELNET and FTP control connections, can be well-modeled as Poisson processes with fixed hourly rates. However, other connection arrivals, such as FTPDATA connections, exhibit significant burstiness and do not follow Poisson distributions. TELNET packet interarrivals, when modeled using exponential distributions, severely underestimate burstiness, while empirical Tcplib distributions better reflect actual traffic patterns. FTPDATA connections within FTP sessions are clustered into "bursts," with the largest bursts dominating FTPDATA traffic. The authors also find that wide-area traffic exhibits self-similarity, with burstiness persisting over many time scales. This has implications for congestion control and traffic performance. The paper concludes that self-similar models may better capture the characteristics of wide-area traffic than Poisson processes.