Cloud providers should ensure QoS while maximizing resources utilization. One optimal strategy is to timely allocate resources in a fine-grained mode according to application's actual resources demand. The necessary precondition of this strategy is obtaining future load information in advance. We propose a multi-step-ahead load forecasting method, KSwSVR, based on statistical learning theory which is suitable for the complex and dynamic characteristics of the cloud computing environment. It integrates an improved support vector regression algorithm and Kalman smoother. Public trace data taken from multitypes of resources were used to verify its prediction accuracy, stability, and adaptability, comparing with AR, BPNN, and standard SVR. Subsequently, based on the predicted results, a simple and efficient strategy is proposed for resource provisioning. CPU allocation experiment indicated it can effectively reduce resources consumption while meeting service level agreements requirements.

No abstract

No abstract

In order to improve the host energy efficiency in IaaS, we proposed an adaptive host resource provisioning method, CoST, which is based on QoS differentiation and VM resizing. The control model can adaptively adjust control parameters according to real time application performance, in order to cope with changes in load. CoST takes advantage of the fact that different types of applications have different sensitivity degrees to performance and cost. It places two different types of VMs on the same host and dynamically adjusts their sizes based on the load forecasting and QoS feedback. It not only guarantees the performance defined in SLA, but also keeps the host running in energy-efficient state. Real Google cluster trace and host power data are used to evaluate the proposed method. Experimental results show that CoST can provide performance-sensitive application with a steady QoS and simultaneously speed up the overall processing of performance-tolerant application by 20~66%. The host energy efficiency is significantly improved by 7~23%.

Aimed at the vehicle/pedestrian visual sensing task under low-light conditions and the problems of small, dense objects and line-of-sight occlusion, a nighttime vehicle/pedestrian detection method was proposed. First, a vehicle/pedestrian detection algorithm was designed based on You Only Look Once X (YOLOX). The model structure was re-parameterized and lightened, and a coordinate-based attention mechanism was introduced into the backbone network to enhance the feature extraction efficiency of vehicle/pedestrian targets. A feature-scale fusion detection branch was added to the feature pyramid, while a loss function was designed, which combines Complete Intersection Over Union (CIoU) for target localization and Varifocal Loss for confidence prediction to improve the feature extraction ability for small, dense, and low-illumination targets. In addition, in order to further improve the detection accuracy of the algorithm under low-light conditions, a training strategy based on data domain transfer was proposed, which fuses the larger-scale daylight dataset with the smaller-scale nighttime dataset after low-illumination degrading. After low-light enhancement, training and testing were performed accordingly. The experimental results show that, compared with the original YOLOX model, the improved algorithm trained by the proposed data domain transfer strategy achieved better performance, and the mean Average Precision (mAP) increased by 5.9% to 82.4%. This research provided effective technical support for autonomous driving safety at night.