Empirical evaluation of upstream throughput in a DOCSIS access network

We present empirical measurements of the upstream throughput of a DOCSIS 1.1 link. In contrast to all previous simulation-based studies, our measurements have been obtained from actual cable-modems (CMs) and head-ends, each from two different vendors. We have constructed an exhaustive database of measurements of a large subset of the space of parameters affecting upstream throughput. Using a well-known non-parametric hypothesis test, we query this database for obtaining statistically robust answers to key questions about the effect of parameter changes on the throughput. Our results indicate that, for a single CM scenario, packet concatenation is most effective, whereas piggybacking is effective and better than concatenation only in some cases. Using both enhancers decreases throughput for a single CM scenario. Our results are robust across head-end implementations and are of immediate interest to network and protocol architects as well as device developers.