AFTER: Algorithmic Framework for Throughput EstimatoRs for IEEE 802.11 networks

Predicting performance in real-time in multihop wireless networks is still a challenge. This problem arises, for example, in routing decisions and QoS provisioning. The literature includes a number of prediction techniques, but usually they are based on simplifications or information that lowly correlates with actual performance. More complex mathematical models exist, but they are computationally expensive and not as accurate as one might expect. An alternative is the usage of network simulation. However, traditional stochastic network simulators require high levels of user intervention and long execution times for reliable results due to their randomized nature. Thus, they are not suitable for autonomic applications that demand real-time responses. This work proposes AFTER (Algorithmic Framework for Throughput EstimatoRs), an on-the-fly algorithm for real-time throughput estimation for multihop IEEE 802.11 based networks. Instead of mathematical models that are either too complex or overly simplified, AFTER uses a simulation-based approach. Differently from stochastic network simulators, AFTER replaces random events for their expected outcome. This procedure continues until AFTER is able to determine the long term average throughput for all network flows. We evaluate AFTER using a number of multihop wireless scenarios and results show that it accomplishes its goals. In comparison to stochastic network simulators, AFTER is able to greatly reduce the time needed for predicting throughput and completely eliminates the need for user intervention. On the other hand, in comparison to mathematical models, AFTER achieves better accuracy and correlation with the actual throughput. Moreover, due to its simulation-based nature, AFTER can be extended to incorporate complex network interactions.