Summary With the increase of data processing and Hadoop data center construction requirements, the performance of Hadoop data center is limited by inappropriate resources utilization. This paper introduces a new method to predict utilization for large‐scale Hadoop clusters. The new method adopts a two steps model, which includes Hadoop applications' performance simulation and resources utilization prediction. For performance simulation, a new simulator, which integrates baseline test and multilayered network model, is introduced and implemented. A resources utilization predictor is proposed in the second step. By analyzing the pattern of resources utilization, a single task model is proposed. A parallel‐batch‐task‐based (PBT) model, which represents the behavior of real Hadoop applications by integrating the single task model, is introduced. Two test scenarios are configured to verify the performance of our method. For the data center scenario, Terasort, Wordcount, and Hive are selected as benchmarks. In the virtual machines scenario, Terasort is used as benchmark. The experiments show that the error comparing between the simulator results and experimental environment results in most cases is less than 10%. The results confirm that we can locate the resource bottleneck for Hadoop clusters, meanwhile we can agilely configure clusters for applications with massive data.

In this work, we address the task of dense stereo matching with Convolutional Neural Networks (CNNs). Particularly, we focus on improving matching cost computation by better aggregating contextual information. Towards this goal, we advocate to use atrous convolution, a powerful tool for dense prediction task that allows us to control the resolution at which feature responses are computed within CNNs and to enlarge the receptive field of the network without losing image resolution and requiring learning extra parameters. Aiming to improve the performance of atrous convolution, we propose different frameworks for further boosting performance. We evaluate our models on KITTI 2015 benchmark, the result shows that we achieve on-par performance with fewer post-processing methods applied.

Multi-core processors achieve a trade-off between the performance and the power consumption by using Dynamic Voltage Scaling (DVS) techniques. In this paper, we study the power efficient scheduling problem of real-time tasks in an identical multi-core system, and present Node Scaling model to achieve power-aware scheduling. We prove that there is a bound speed which results in the minimal power consumption for a given task set, and the maximal value of task utilization,$u_{max}$, in a task set is a key element to decide its minimal power consumption. Based on the value$u_{max}$, we classify task sets into two categories: the bounded task sets and the non-bounded task sets, and we prove the lower bound of power consumption for each type of task set. Simulations based on Intel Xeon X5550 and PXA270 processors show Node Scaling model can achieve power efficient scheduling by applying to existing algorithms such as EDF-FF and SPA2. The ratio of power reduction depends on the multi-core processor's property which is defined as the ratio of the bound speed to the maximal speed of the cores. When the ratio of speeds decreases, the ratio of power reduction increases for all the power efficient algorithms.