Abstract. Mobile phones are set to become the universal interface to online services and cloud computing applications. However, using them for this purpose today is limited to two configurations: applications either run on the phone or run on the server and are remotely accessed by the phone. These two options do not allow for a customized and flexible service interaction, limiting the possibilities for performance optimization as well. In this paper we present a middleware platform that can automatically distribute different layers of an application between the phone and the server, and optimize a variety of objective functions (latency, data transferred, cost, etc.). Our approach builds on existing technology for distributed module management and does not require new infrastructures. In the paper we discuss how to model applications as a consumption graph, and how to process it with a number of novel algorithms to find the optimal distribution of the application modules. The application is then dynamically deployed on the phone in an efficient and transparent manner. We have tested and validated our approach with extensive experiments and with two different applications. The results indicate that the techniques we propose can significantly optimize the performance of cloud applications when used from mobile phones.

No abstract

With the functionality of mobile applications ever increasing, designers are often confronted with either the resource limitations of the devices or of the network. As pointed out by recent work, application partitioning between mobile devices and clouds, can be used to solve some of these issues, improving performance and/or battery life. In this paper, we argue that the static decisions made in existing work cannot leverage the full potential of application partitioning. Thus, to allow for variations in the execution environment, we have developed a system that dynamically adapts the application partition decisions. The system works by continuously profiling an applications performance and dynamically updating its distributed deployment to accommodate changes in the network bandwidth, devices CPU utilization, and data loads. Using several real applications, we show that our approach provides performance gains as high as 75% over traditional approaches and achieves lower power consumption by a factor close to 45%.

In this work we present MTEX-CNN, a novel explainable convolutional neural network architecture which can not only be used for making predictions based on multivariate time series data, but also for explaining these predictions. The network architecture consists of two stages and utilizes particular kernel sizes. This allows us to apply gradient based methods for generating saliency maps for both the time dimension and the features. The first stage of the architecture explains which features are most significant to the predictions, while the second stage explains which time segments are the most significant. We validate our approach on two use cases, namely to predict rare server outages in the wild, as well as the average energy production of photovoltaic power plants based on a benchmark data set. We show that our explanations shed light over what the model has learned. We validate this by retraining the network using the most significant features extracted from the explanations and retaining similar performance to training with the full set of features.

No abstract