The metrics that will be collected for the TCP Layer include Packet Error
Ratio, Packet Interarrival Time, and Packet Size.
Table 
summarizes the scenarios in which these measurements
will take place.
Again, the packet size is clearly topology-independent, so it will be determined using the KU/TIOC/NRL three-point configuration.
In the TCP layer, we define the Packet Error Ratio as:
Considerations on lost and misinserted packets are left to the lower layers, as we will see in the following sections. This metric is not highly dependent on the network load; for this reason, we will run the experiment on only a few representative topologies.
On the other hand, the Packet Interarrival Time is highly dependent on the test topology. As the load on a particular segment of the network is increased, we expect to see a higher variance on the Interarrival Time. Among our chosen topologies, the highest variance should occur for the All-Site Mesh Configuration.
We should also bear in mind that the Packet Interarrival Time can be interpreted as an indication of delay jitter. If each packet is timestamped at the transmitter and we record the arrival times at the receiver, we can calculate the delay variance, independently of transmitter-receiver synchronization.
Let's suppose n packets are transmitted at times 
and received at times 
.
We can calculate the ``relative delay'' 
as:
Note that when the receiver and the transmitter are synchronized this relative delay is equal to the packet delay.
Then, the variance of the relative delay is a measure of the delay jitter. Notice that for this measurement of delay jitter, it is not necessary that the transmitter and receiver clocks be synchronized; therefore, no synchronization techniques, such as the Network Time Protocol (NTP), are assumed.
