Search

Video Analytics System has emerged as a promising technology to realize deep neural network based intelligent applications for video streams. Its objective is to maximize the video analytics performance of video streams, such as accuracy, while utilizing the limited computing resource capacity efficiently. Existing video analytics systems attempt to adapt configurations to optimize resource-accuracy tradeoffs under limited resource capacity. Especially, several works recently propose a configuration adaptation algorithm based on online profiling to overcome the inefficiency of one-time offline profiling caused by the dynamics of the configuration's impact on video analytics accuracy. However, their systems are inefficient or limited to process the video analytics of multiple video streams in a GPU-enabled edge server. Furthermore, their online profiling methods still leads to a high profiling cost. In this paper, we design a video analytics system to adapt configurations for optimizing the resource-accuracy tradeoffs of multiple video streams with respect to frame rate and resolution fully under limited resource capacity of a GPU-enable edge server. In addition, we propose a lightweight online profiling method, utilizing the underlying characteristics of the video objects. Then, based on it, we propose a configuration adaptation algorithm to find the best configuration of each video stream and minimize accuracy degradation of multiple video streams under limited resource capacity of a GPU-enable edge server. To evaluate the proposed algorithm, we use a subset of surveillance videos and annotations from VIRAT 2.0 Ground dataset. The experimental results show that, in a GPU-enabled edge server, our video analytics system achieves the optimal configuration adaptation on resource-accuracy tradeoffs and our algorithm reduce the profiling cost of existing systems significantly while achieving the video analytics performance comparable to them.

Cloud-edge-collaborative storage (CECS) is a promising framework to process data of the internet of things (IoT). It allows edge servers to process IoT data in real-time and stores them on a cloud server. Hence, it can rapidly respond to the requests of IoT devices, provide a massive volume of cloud storage for IoT data, and conveniently share IoT data with users. However, due to the vulnerability of edge and cloud servers, CECS suffers from the risk of data leakage. Existing secure CECS schemes are secure only if all edge servers are trusted. In other words, if any edge server is compromised, all cloud data (generated by IoT devices) will be leaked. Additionally, it is costly to request expected data from the cloud, which is linear with respect to the number of edge servers. To address the above problems, we propose a new secure data search and sharing scheme for CECS. Our scheme improves the existing secure CECS scheme in the following two ways. First, it enables users to generate a public-and-private key pair and manage private keys by themselves. In contrast, the existing solution requires edge servers to manage users' private keys. Second, it uses searchable public-key encryption to achieve more secure, efficient, and flexible data searching. In terms of security, our scheme ensures the confidentiality of cloud data and secure data sharing and searching and avoids a single point of breakthrough. In terms of performance, the experimental results show that our scheme significantly reduces users' computing costs by delegating most of the cryptographic operations to edge servers. Especially, our scheme reduces the computing and communication overhead for generating a search trapdoor compared with the existing secure CECS scheme.

With the growing volumes of data, many organizations are deploying geo-distributed edge servers and building federations atop of edge servers to improve data sharing, effective collaborations, and analytics. Multiple federation file systems are designed to satisfy such needs, but due to application-specific architectures, these federations neglect some important features that can improve the overall federation performance. In this paper, we address the important challenges of federated file systems, in particular, global namespace, optimal data placement and analysis, efficient data migration across edge servers, and metadata optimizations. To further investigate these challenges, we prototyped the federation file system iStore to emulate the federation and showed the significance of the afore-mentioned key challenges in the federation. The iStore provides unified global namespace atop of geo-distributed edge servers with a generic job and resource-aware data storage and placement algorithm (JRAP), which minimizes the job execution time by considering resources at each edge server. Furthermore, to enable effective data migration, we employed direct channel file layout-aware data transfer and designed a batch-based metadata scheme for federations to reduce the metadata contention with increasing clients. We evaluated the efficacy of various big data applications from data generation to storage and analysis using the iStore on real testbed and simulation.

Deep neural networks (DNNs) have been extensively used in multiple applications. Due to the over-parameterized DNN-based model, the mobile device has computation and energy limitations to deploy such a model for machine learning tasks. Thus, many works focus on compressing a large-scale model to a small-scale model. In addition, the mobile device training a model assisted by the edge server is an emerging solution in the edge computing environment. However, recent researches have found that DNNs are vulnerable to adversarial examples. These crafted adversarial examples can fool a DNN-based model incorrect predictions. In particular, the DNN-based model can cause risks when used in safety-critical settings. To address this problem, we design a framework for generating a robust deep-convolutional-neural-network-based compressed model in the edge computing environment. The model is partitioned and trained by the mobile device and the edge server. The robust compressed model is constructed mainly via model compression and model robustness. In model robustness, a defensive mechanism is proposed for enhancing the robustness of the compressed model against adversarial examples. Furthermore, the weight distribution of the compressed model is considered for improving the model's accuracy in the defense method. The small-scale compressed model is effective and robust as a collaborative device-server inference for providing recognition tasks for the near devices. On the other hand, it is practical to deploy on the mobile device due to its small-size. Experimental results show that the generated compressed model has strong robustness against adversarial examples while holding high accuracy.

Collaboration spaces formed from edge servers can efficiently improve the quality of experience of service subscribers. In this paper, we first utilize a strategy based on the density of Internet of Things (IoT) devices and ${k}$ -means algorithm to partition network of edge servers, then an algorithm for IoT devices’ computation offloading decisions is proposed, i.e., whether we need to offload IoT devices’ workload to edge servers, and which edge server to choose if migration is needed. The combination of locations of edge servers and the geographic distribution of various IoT devices can significantly improve the scheduling of network resources and satisfy requirements of service subscribers. We analyze and build mathematical models about whether/how to offload tasks from various IoT devices to edge servers. In order to better simulate operations of the mobile edge servers in more realistic scenarios, the input size of each IoT device is uncertain and regarded as a random variable following some probability distribution based on long-term observations. On the basis of that, an algorithm utilizing sample average approximation method is proposed to discuss whether the tasks to be executed locally or offloaded. Besides, the algorithm proposed can also help decide whether service relocation/migration is needed or not. Finally, simulation results show that our algorithm can achieve 20% of global cost less than the benchmark on a true base station dataset of Hangzhou.