Are Smaller Packets Less Likely to Be Lost?

摘要

Despite growth in internet link capacities and the efforts of traffic engineers, packet loss continues to be a reality in today's internet. While many tools for active measurement of packet loss have been developed over the years, applying these tools to a particular operational network context in order to most accurately estimate loss characteristics has remained a challenge. In this paper we examine the problem of selecting a probe packet size for active loss measurement. Prior work has used either small (e.g., 40 byte) packets in order to minimize probing bandwidth, or larger (e.g., 1000 byte) packets in order to increase the likelihood that a probe will be lost if congestion is taking place along a path. We examine these rules of thumb in a systematic way through a series of controlled laboratory experiments. We find that a-perhaps the-key system parameter is whether network queue limits are specified in terms of packets or bytes. Our results suggest that if queue limits are specified in terms of packets, there may be no clear optimal probe packet size that yields the most accurate results. On the other hand, if queue limits are specified in bytes, a packet size chosen based on the average packet size of aggregate background traffic leads to the most accurate loss average estimate.