The observations from the experiments allow us to draw the following general conclusions about the network behavior as seen on the connections studied:
1) The loss characteristics of a connection seem to change significantly during a day and from one day to the next. Even for the runs conducted at early morning hours when we expect the traffic to be light, the number of losses to most sites was non-negligible. Clearly the traffic to overseas sites may not have light traffic at early morning hours according to the Eastern Standard Time in the United States.
2) The round trip time, RTT, is clearly affected by the network traffic. It is interesting to note that with the fine granularity of the observations (40 ms) the RTT varies widely over short periods of time. The range of variability, however, remains within reasonable bounds over several seconds.
3) For every connection, we observed that the value of RTT increased very sharply and attained a value which is much higher than the normal range of variability. For many connections, such sharp rises occurred periodically. For example, the connection to Bethesda shows a pronounced periodic RTT peaks, which are about 750 packets apart.
4) There were no instances of duplicate packets normally, except for a few runs to Italy. The number of duplicate packets ranged from 1 to 2. We did observe reordering of packets for most connections. However, it seems that the number of reorders is uncorrelated with the number of losses.
5) While many loss occurrences are preceded or followed by a packet received with high RTT value, there are instances of losses that are preceded or followed by packets received with nominal RTT values. If the cause of the packet loss is buffer overflow, and the buffer can hold packets received over a period longer than 40ms, we expect that the packet before and/or packet after an isolated loss to have a higher RTT. Thus, it seems that there may be some phenomenon other than the buffer overflow that may be causing packet losses. There is a possibility that there are some locations with very small buffers that are discarding packets under some conditions.
6) As observed through the use of traceroute outputs, the path remained mostly the same from the source to the destination host for a given link. In some cases however, the route did change dynamically. The most significant change in route was observed in one of the experiment runs to Italy, whereby hops between router number 6 and router number 10 were replaced by 9 hops, which increased the total number of hops to 22 from 18.
