Unreliable Transport (UDP)

  In this section we present measurements of throughput and jitter for UDP traffic through the Ricochet system.

Throughput: UDP traffic is one-way and does not incur reverse channel contention as TCP does. Thus, the half-duplex nature of the packet radios does not affect UDP throughput. Our measurements show a maximum throughput of 50-58 kb/s for UDP traffic, across a range of packet sizes. This is approximately double the maximum throughput of TCP traffic, and triple the TCP throughput of smaller packet sizes. The performance of simultaneous rate-controlled UDP transfers degrades almost linearly - our results for two parallel UDP streams show each obtaining 25 kb/s, and for three parallel streams each obtaining about 16 kb/s. This matches the hypothesis about the cause of degradation in performance in the bi-directional TCP traffic versus the uni-directional UDP traffic.

Jitter: The jitter of a packet stream is defined as the mean deviation of the difference in packet spacing at the receiver compared to the sender, for a pair of packets. More formally, if is the time packet was sent from the sender, and is the time it was received by the receiver, the jitter sample is given by:

and the average jitter for is simply the average value over n packets.

The jitter of a stream is a particularly important quantity for time sensitive data such as real-time audio and video, since a large jitter can have a pronounced effect on the perceived quality.

In order to confine the jitter measurements to the Ricochet system, we measured the jitter solely on the experimental Ricochet network. Figure 5 shows a plot of the average jitter as a function of the transmission rate for a packet size of 1024 bytes. We see that higher transmission rates cause the jitter to increase. The reason for this is due to the fact that the increased transmission rate causes the queues in the MGW and intermediate packet radios to build up, causing irregularities in the packet inter-spacing. This in turn translates to increased average jitter. We see although that negligible levels of jitter can only be obtained at data rates below 40 kb/s, the maximum jitter we see is only approximately 30 msec which is relatively low. As a guideline, 30 msec matches a single frame time in a full-motion video stream, or a little over the acceptable level of jitter for an audio stream. However, we also note that these measurements are ``best-case'' measurements under no other external load, and in particular, under no reverse channel traffic from the mobile host. A more complete analysis of jitter and UDP performance in the presence other traffic is a topic for future work.