Average one-way delay

Comments:
  1. One-way delay increases linearly with the packet size both with and without best-effort traffic;
  2. Without best-effort traffic the PB size does not impact the average latency as expected, since in these case a premium packet is serviced by the scheduler as soon as it arrives in the queue, so the premium queue never builds up. For this reason, the queue depth is irrelevant since the queue is always almost empty.
    With best-effort traffic the average latency is again constant for a given packet size, with the only exception of when the PB size is of 1500 bytes: in this case, for any packet size the average latency is approximately 1.8 msec larger.
  3. The slope of graph representing the average rate is the same with and without best-effort traffic.
  4. As expected, when best-effort traffic is added, the average delay increases of 1.7 msec ( figure 3). An explanation of this could be that when a premium packet is dequeued a best-effort packet could be under transmission in the tx queue. However, the additional delay is less than the transmission time of a best-effort packet (4 msec for 1000 bytes at a line rate of 2 Mbps), in fact the one-way delay plotted in figure 2 is the average.
  5. The exception case for 1500 bytes with best-effort traffic (figure 2) is under investigation.
Figure 1: average one-way delay versus packet size for different premium buffer sizes, without best-effort traffic.
Figure 2: average one-way delay versus packet size for different premium buffer sizes, with best-effort traffic. Delay is the same independently of the buffer size, with the only exception of the case where PB = 1500 bytes.
Figure 3: comparison of one-way delay with and without best-effort traffic.

Last modified: Nov 22, 1999