The paper "Predictive Resource Management for Wearable Computing" by Dushyanth Narayanan and M. Satyanarayanan addresses the challenge of achieving crisp interactive response in resource-intensive applications on resource-poor wearable hardware. The authors propose a solution based on multi-fidelity computation supported by predictive resource management, which can significantly reduce both the mean and variance of response time. They demonstrate a 60% reduction in mean latency and a 30% reduction in the coefficient of variation for augmented reality applications. The key to this performance improvement is a history-based approach to demand prediction. The paper introduces the concept of multi-fidelity computation, where applications can present results at different fidelities, allowing users to trade off output quality for lower resource consumption. The system, called Odyssey, automatically makes runtime fidelity decisions on behalf of the applications, freeing programmers from this burden. Odyssey uses history-based prediction to estimate resource demand and performance, and a utility function to capture user preferences. The authors evaluate their approach using four applications, focusing on augmented reality as a motivating example. They show that history-based demand prediction is feasible, accurate, and necessary for performance improvement. The implementation of Odyssey is based on the Odyssey system, which originally supported fidelity for stored data. The paper also discusses the design rationale, the programming interface, and the system architecture. The evaluation section validates the approach by answering three sets of questions: the accuracy of history-based demand prediction, the performance benefits of predictive resource management, and the programming costs and runtime overheads. The results show that Odyssey can reduce mean latency by 60% and variability by 30% for augmented reality applications, with prediction errors as low as 0.3% for some resources. The cost of using Odyssey is modest, involving additional lines of code and a small runtime overhead. The paper concludes by discussing future research directions, including testing with full-fledged AR applications, automating the construction of demand predictors, combining demand prediction with QoS-based allocation, and exploring mixed-initiative approaches.The paper "Predictive Resource Management for Wearable Computing" by Dushyanth Narayanan and M. Satyanarayanan addresses the challenge of achieving crisp interactive response in resource-intensive applications on resource-poor wearable hardware. The authors propose a solution based on multi-fidelity computation supported by predictive resource management, which can significantly reduce both the mean and variance of response time. They demonstrate a 60% reduction in mean latency and a 30% reduction in the coefficient of variation for augmented reality applications. The key to this performance improvement is a history-based approach to demand prediction. The paper introduces the concept of multi-fidelity computation, where applications can present results at different fidelities, allowing users to trade off output quality for lower resource consumption. The system, called Odyssey, automatically makes runtime fidelity decisions on behalf of the applications, freeing programmers from this burden. Odyssey uses history-based prediction to estimate resource demand and performance, and a utility function to capture user preferences. The authors evaluate their approach using four applications, focusing on augmented reality as a motivating example. They show that history-based demand prediction is feasible, accurate, and necessary for performance improvement. The implementation of Odyssey is based on the Odyssey system, which originally supported fidelity for stored data. The paper also discusses the design rationale, the programming interface, and the system architecture. The evaluation section validates the approach by answering three sets of questions: the accuracy of history-based demand prediction, the performance benefits of predictive resource management, and the programming costs and runtime overheads. The results show that Odyssey can reduce mean latency by 60% and variability by 30% for augmented reality applications, with prediction errors as low as 0.3% for some resources. The cost of using Odyssey is modest, involving additional lines of code and a small runtime overhead. The paper concludes by discussing future research directions, including testing with full-fledged AR applications, automating the construction of demand predictors, combining demand prediction with QoS-based allocation, and exploring mixed-initiative approaches.