Last time, we discussed why throwing bandwidth at a problem does nothing to improve application performance. We used an example of a company that has a T-1 circuit from Boston to San Francisco that was only getting 56K bit/sec of throughput. We also noted that increasing the circuit to T-3 speed only increases the throughput to 58K bit/sec. Some of the key parameters of this network are that the WAN has 70 milliseconds of one-way delay and that due to the TCP window size, the company sends 4,096 bits before it has to stop and wait for an acknowledgement.
When analyzing this situation, it is helpful to look at two metrics. First is the one-way delay of 70 milliseconds. If the Internet provided the WAN connectivity there would be multiple factors contributing to this delay. But given that a private line provides the connectivity, that 70-millisecond delay is entirely attributable to the speed of light.
The second metric is how long after the first bit arrives in San Francisco, does the last bit arrive. The answer to this (2.7 milliseconds) is calculated by taking the number of bits being sent (4,096) and dividing that by the line speed.
Putting these two metrics together says that the company can transmit 4,096 bits once every 72.7 milliseconds. It also says that no matter how large the pipe is the company can never send more than 4,096 bits every 70 milliseconds. This translates into a maximum throughput of just over 58K bit/sec.
Future newsletters will detail what can be done to improve the performance of this WAN.