Cloud Computing and Storage Laboratory of Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences
Cloud Computing and Storage Laboratory is affiliated to the Cloud Computing Center of the Digital Institute of Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, and is committed to cutting-edge technologies including cloud computing, virtualization, big data analysis and processing, storage and other fields, and has accumulated rich technical achievements and experimental data. CCSLab is led by researcher Wang Yang, with 1 postdoctoral fellow, 2 doctors, several masters of the university, and a number of internship students from universities such as the University of Science and Technology of China. Graduates go to the University of Rochester and well-known domestic Internet companies such as Ali, Tencent, Huawei, and Baidu.
Research directions
Security and acceleration technologies for system virtualization in cloud environments
As an extremely important resource in the data center, the network is the key to restricting the performance of cloud computing systems. Starting from the underlying mechanism of virtualization technology, we study network I/O virtualization and online migration technology, and realize performance improvement under the cloud platform by dynamically sensing and scheduling network loads and applying new high-speed network technologies such as RDMA. At the same time, the cloud environment poses a more severe challenge to the security mechanism, and we use SDN thinking to separate the data path and control path, and also take into account network performance and security through the monitoring of the control path on the basis of virtualized RDMA.
Edge-cloud collaboration technologies and methods for machine learning
Limited by the computing power and network bandwidth of edge devices, computing loads such as machine learning (mainly DNN) are difficult to be widely applied in edge networks. approach to address and optimize the performance of edge intelligence. By dynamically partitioning the DNN, we deploy it between the terminals and edges of the multi-access network to accelerate the training and reasoning tasks, and propose a two-stage decision-making model, and use the ideas of game and reinforcement learning to design algorithms to solve the problem. Optimization problems for DNN dynamic partitioning and service matching.
Blockchain technology
The team is committed to in-depth analysis of the characteristics of the existing consensus mechanism of blockchain (mainly alliance chain), breaking through key technologies such as network latency, security and throughput. Design consensus for different application scenarios to solve the underlying problems of blockchain applications. At the same time, in the human-machine-object fusion system and autonomous driving system, the application of blockchain technology is studied to solve the problem of data storage and sharing, and the data storage that can be safely verified is introduced through blockchain. Improve the ability to collect data from complex environments while verifying its accuracy.
Big data framework optimization method based on new storage and network technology
This paper mainly studies the efficient service implementation mechanism based on cloud-native microservice architecture that supports end-, edge, and cloud AI collaboration. Realize application-aware resource allocation and situational guidance for application load on edge cloud servers, use artificial intelligence algorithms to generate rules and establish rule sets to generate scheduling strategies, and develop a data-driven adaptive self-evolution execution platform. Customize multiple core functions of the platform as microservices, containerized packaging using Docker, and dynamic management through Kubernetes, using centralized orchestration and scheduling algorithms to dynamically manage and schedule resources.
Representative projects
Human-machine-object integration cloud computing architecture and platform data-driven application adaptation and self-evolution technology（National Key R&D Program "Cloud Computing Big Data" Special Project, 2018--2021）
This project solves the three major problems and challenges of multi-dimensional programmability of cloud-network resources, dynamic discovery of cloud-network resources, and trusted collaboration of cloud-network resources, and is the bottleneck of human-machine-object integration as a new generation of information technology. This project proposes a software-defined method for human-machine-object integration, presents a set of software-defined infrastructure for human-machine-object integration, and establishes the basic theory, method and related technical system and specifications of software-defined for human-machine-object integration. Main research content:
（1）Based on the software-defined method, a cloud-network resource collaborative management architecture and performance optimization method for application adaptation are constructed.
（2）Sense and collect resource usage data from specific scenario applications; With the help of fusion and analysis methods, establish dynamic characteristics and prediction mechanisms of application resource usage.
（3）According to the dynamic operating environment and variable user requirements in specific scenarios, a distributed directory service is built to realize the tracking and management of mobile devices, and the computing technology based on the collaborative execution of cloud, network and end is completed by means of task migration and offloading.
Research on key technologies and platforms of unmanned driving driven by collaborative intelligence（Macao Key R&D Special Program , 2017-01--2020-12）
At present, the core functional requirements of multi-vehicle intelligent collaboration include: interconnection of any vehicle, any time and any place; Full-time and spatial dynamic traffic information collection and fusion; Effective coordination of people and roads - collaborative safety (divided into active safety and passive safety), collaborative control (divided into active control and passive control). Research content:
（1）Driving scene recognition technology that integrates radar perception and visual perception
（2）Scenario knowledge base construction technology based on knowledge graph and event graph
（3）Driving rule generation and evolution technology based on reinforcement learning
（4）Multi-vehicle collaboration technology based on cognitive architecture, including multi-vehicle formation motion planning, multi-vehicle collision avoidance, and multi-vehicle competition consensus.
Security control and trusted enhanced management system based on large-scale dynamic infrastructure（Research and development in key areas of Guangdong Province , 2018--2021）
This project focuses on the key technologies of network security boundary control:
（1）Research on the intelligent gateway technology based on security detection, automatically detect security problems through the system, and then implement the corresponding security isolation strategy based on the intelligent gateway.
（2）The active security management technology based on concurrent migration of virtual machines is studied, and the strategy of concurrent migration of multiple virtual machines in the case of security problems is designed to improve the overall security and reliability of the system.
Research on data and service migration of mobile cloud computing in spatiotemporal big data environment（National nature on the project , 2017--2020）
Data and service migration is an effective way to reduce latency and network load and improve the quality of mobile cloud services. Faced with the diversification, real-time, and huge number of user access modes of cloud services in the big data environment, the current migration strategies and methods face serious challenges. Objective: To propose an optimization method for data and service migration based on access spatiotemporal estimation, so that the deployment of shared data and services can match and adapt to the access patterns of mobile users, and complete the requested services in the least cost.
Representative papers
Cost-Driven Data Caching in the Cloud: An Analytic Approach, IEEE International Conference on Computer Communications. (IEEE INFOCOM, 2021)
Sova: A Software-Defined Autonomic Framework for Virtual Network Allocations. (IEEE Transactions on Parallel and Distributed Systems, 2021)
Deadlock Avoidance Algorithms for Recursion-Tree Modeled Requests in Parallel Executions. (IEEE Transactions on Computers, 2021)
Algorithmics of Cost-Driven Computation Offloading in the Edge-Cloud Environment. (IEEE Transactions on Computers, 2020)