Description

The delay information provided in this section corresponds to the delay experienced by packets between two points in the Sprint IP backbone network, where IPMON systems have been installed. Packet traces collected at different points in the network are matched, and the common packets are identified. The difference between the GPS timestamps of the common packets provides an accurate measure of the delay experienced by packets through the network.

The results are presented in a few graphs and tables:

• Delay Statistics This table provides the basic statistics of delay derived from matching two traces.
• Delay Distribution This graph shows the percentage of packets that experience a given network delay. For single-hop delays (i.e. delays across one routers), the distribution may present multiple spikes that, in general, are related to the size of the packets. For longer distances (e.g. delays from San Jose to New York City), multiple peaks may instead be due to load balancing happening along the path or to re-routing events. Using the information in the other graphs, we can distinguish between load balancing and re-routing.
• Delays over Time This graph plots packet delays over time. Each dot in the graph represents the minimum, average and maximum delay over a 10s time interval. The minimum delay can give a hint on the cause behind multiple spikes in the delay distribution. If the minimum delay is constant over the entire duration of the traces, then we can easily conjecture that load balancing is taking place over the path. Otherwise, minimum delay changing abruptly in the middle of the trace is a clear indication of traffic being re-routed because of a topology change.
• Matched Packet Count over Time This graph shows the number of matched packets in the pair of traces over time. Each dot represents the number of packets for which we have a delay measure over a 10s interval. This graph provides additional information on the representativeness of our measurements and on the load variation of this path (e.g. between router 1 in San Jose and router 2 in New York City).
• Number of Hops (TTL Difference) In this last table we show the number of router hops crossed by the observed packets. This table may provide additional information to explain multiple peaks in the delay distribution.