This paper presents the Weight-Watcher service. This service aims at providing resource consumption measurements and estimations for software executing on resource-constrained devices. By using the Weight-Watcher, software components can continuously adapt and optimize their quality of service with respect to resource availability. The interface of the service is composed of a Profiler and a Predictor. We present an implementation that is lightweight in terms of CPU and memory. We also performed various experiments that convey (a) the trade-off between the memory consumption of the service and the accuracy of the prediction, as well as (b) a maximum overhead of 10% on the execution speed of the VM for the Profiler to provide accurate measurements.
展开▼
