Data-analytics frameworks such as Apache Hadoop have tunable configuration parameters set by experienced system administrators and/or application developers. However, tuning parameters manually can be hard and time-consuming because it requires domain-specific knowledge and understanding of complex inter-dependencies amongst parameters. Most of the frameworks seek efficient resource management by using slots or containers as resource units to be assigned to jobs or tasks, the maximum number of slots or containers in a system being part of the static configuration of the system. This paper proposes a hierarchical self-tuning approach using a fuzzy-logic controller to dynamically adjust the number of concurrent jobs and additional controllers (one for each cluster node) to adjust the number of resource units assigned to jobs on each node. To manage the maximum number of available resource units in each node, the controllers take resource usage by other processes (e.g., system processes) into account. A prototype of our approach was implemented for Apache Hadoop on a cluster running at CloudLab. The experimental evaluation shows that the proposed approach yields up to a 42% reduction of the jobs makespan that results from using Hadoop default settings.
A major concern for today's smartphones is their much faster battery drain than traditional feature phones. The difference is mainly contributed by those more powerful but also much more power-consuming smartphone components, such as the multi-core application processor and the high-definition (HD) display. In this paper, we investigate how to increase the battery life of smartphones by minimizing the use of application processor and HD display for operations related to basic functions. We find that the application processor is often waken up by a process running on it, called the Radio Interface Layer Daemon (RILD), which interfaces users/apps to the GSM/LTE cellular network. Consequently, we design a Smart On Demand (SOD) configuration that reduces the smartphone energy consumption by running RILD on a secondary low-power microcontroller and by using a secondary low-power display to interface the user with basic functions. Thus, basic phone functions are handled at much lower energy costs while the power-consuming application processor and HD display are waken up only when one needs to use smart apps. We have built a prototype of SOD and evaluated it with real user traces. Our results show that SOD can increase its battery life by up to 2.5 more days.
Inside a logistic system, actors of the logistics have to interact together to manage coherent flow of goods. But they also must deal with the constraints of their environment. The paper's first goal is to study how macro properties (such as global performance) emerge from the dynamic and local behaviors of actors and the structure of the territory. And the second is to understand which local parameters affects these macro properties. A multi-scale approach, made of an agent-based model coupled with dynamic graphs, describes the system's components, such as actors and the transportation network. Adaptive behaviors are implemented in this model (with data about the Seine axis) to highlight the system's dynamics. Agent strategies are evolving according to traffic dynamics and disruptions. This logistic system simulator has the capacity to exhibit large scale evolution of territorial behavior and efficiency face to various scenarios of local agent behaviors.
This work introduces Policy Reuse for Safe Reinforcement Learning (PR-SRL), an algorithm that combines Probabilistic Policy Reuse and teacher advices for safe exploration in dangerous and continuous state and action reinforcement learning tasks. The algorithm uses a progressive risk function which allows to identify the probability to end up in a fail from a given state. Such risk function is defined in terms of how far such state is from the state space known by the learning agent. Probabilistic Policy Reuse is used to safely balance the exploitation of actual learned knowledge, the exploration of new actions and the request of teacher advice in considered dangerous parts of the state space. Specifically, the pi-reuse exploration strategy is used. Using experiments in the helicopter hover task and a business management problem, we show that the pi-reuse exploration strategy can be used to completely avoid the visit to undesirable situations, while maintaining the performance (in terms of the classical long-term accumulated reward) of the final policy achieved.