When Social Perception meets Edge Computing: Vision and Challenge
0. The Abstract in this paper
This article defines SSEC driven by some fundamental technology trends, presents several examples, analyzes key challenges, and predicts future directions
Keywords: social perception, edge computing, Internet of Things, smart city
I. the INTRODUCTION is introduced
What is social sensing? The data is derived from people or surrounding (perceptual) devices and analyzed for practical purposes such as traffic management, disaster prediction, health management, etc. At present, the processing of social awareness data is carried out in the cloud server, but in fact, the computing capacity of the current edge devices is gradually enhanced, and complex computing can be carried out on the edge devices. Edge devices include mobile phones, tablets, smart wearable devices, various Internet of Things devices, etc.
Figure 1 illustrates the joint (SSEC) structure of social awareness and edge computing:
1) At the edge layer, mobile phones or other smart devices are used as edge devices to collect, store and process data;
2) Edge service layer, edge server (local server, mobile cloud, intelligent routing, gateway) provides additional local storage and computing services, which is located between edge layer and service layer;
3) The service layer, which provides global services to all applications/users using the service.
The advantages of SSEC are:
1) Processing perceptual data directly at the edge can reduce the communication cost (bandwidth) and improve the quality of service (delay);
2) Edge devices can get rewards when they use idle resources to perform tasks;
3) SSEC structure is not affected by single point of failure and can solve back-end bottlenecks.
However, SSEC also has many challenges, and SSEC brings a new set of challenges to real-time resource management by enabling delay-sensitive social sensing applications in edge computing systems. Edge devices are often very “selfish” and are generally not interested in performing sensing tasks and sharing private devices unless there is a very high return. This is quite different from traditional cloud solutions that require high sharing. SSEC requires that edge devices fully share information and write to each other, but edge devices often cannot fully trust each other because they are “selfish” and afraid of security attacks.
The rest of the paper defines SSEC, proposes technologies and implications, discusses some important applications and cases, points out research challenges, and finally draws a blueprint.
II. SOCIAL SENSING EDGE COMPUTING PARADIGM
Iot devices generate huge amounts of data, forcing computations at the edge. Some previous researches, such as microdata center and mobile cloud, have solved part of the problems of cloud computing, but they have not made use of private edge devices like edge computing, and still rely on the configuration of infrastructure.
A. What is SSEC? What is SSEC?
Definition 1. Edge computing based on social Awareness (SSEC) : An application paradigm that uses people and surrounding devices to sense, process, and analyze relevant data collected by the physical world.
In this definition, personal devices not only collect data, but also actively participate in computing and analysis tasks in applications. These devices are diverse, including but not limited to GPS sensors, robots, and multiprocessor servers.
SSEC has two important characteristics: 1, people-oriented; 2. Ability to support various applications of different system structures.
1. People-oriented: 1) The most core concerns of users must be considered: privacy and security, compliance and churn, and rewards (Part IV). 2) Not only devices can participate in perception and computing tasks, but also people can participate directly. In fact, many social awareness apps require direct input. People are nodes on the social edge, and they can use data to draw inferences. This is a special feature of SSEC
2, support the flexibility of various systems:
Hierarchical architecture: Cloud servers have massive storage capabilities, management capabilities, and interfaces for users. The application manages a series of spatially distributed edge clusters, with one edge server and many edge devices in one cluster. The key characteristic of hierarchical architecture is that data flow is fixed: edge devices process data locally and unload it to edge servers, which further process the data and transmit the results to cloud servers for data fusion and storage.
Collaborative Edge: Devices near each other form a self-organizing cluster. Devices in the cluster provide P2P services. This structure is appropriate when the edge or cloud server is not fully available, or to avoid recurrent overhead.
Hybrid mode: A combination of vertical mode and edge cooperative mode. Edge devices are connected to available infrastructures that provide services when edge self-organizing devices cannot meet QoS (quality of service) requirements.
B: Why We Need SSEC
Driven by a number of key technology trends: 1) The computing power of some iot devices is becoming more powerful, even comparable to that of servers in traditional edge computing systems. 2) The popularity of mobile payments provides a more convenient way to use idle resources to complete social awareness tasks and obtain rewards.
- SSEC’s advantages:
1) Coverage and availability: Edge devices around the world can be mobile phones and data processing, coverage is obviously due to the traditional fixed technology infrastructure network. Edge devices also provide mobility, when resource availability becomes closely related to noteworthy events.
2) Delay reduction: The perceptual information already in the device or collected by the device can be directly processed, thus greatly reducing the communication cost (bandwidth) and improving the quality of service (reducing delay). So SSEC is good for real-time applications or time sensitive applications.
3) Practicability: Compared with traditional upload tasks, SSEC takes full advantage of the perception and computing power of edge devices. Computing at the edge eliminates single points of failure and alleviates problems with upload servers (deployment costs of aware tasks and back-end infrastructure costs).
4) Earn returns: Participants can earn returns by contributing local device resources to perform SSEC application computing tasks, and can improve the utilization of idle resources.
III. Examples of real-world APPLICATIONS
Disaster and Emergency Response
The market for sensors on citizens, first responders or journalists spontaneously reports sensory information about a series of unfolding events. SSEC offers a suitable architecture for such applications: 1) Edge devices close enough to body sensors to collect and extract useful characteristic information to the cloud (rather than sending all the data). 2) The edge server layer can collect the processed data to provide real-time updates for residents. 3) The cloud server integrates all the collected information and provides it to relevant organizations or the general public.
Figure 3 is an example. Mobile phones, on-board sensors, cameras, etc., provide the first time images to the server, and these data deduce the suspect’s escape route. Edge servers can also provide real-time alerts of potential hazards and provide security advice.
B. Collaborative Traffic Monitoring
Collaborative traffic monitoring in social awareness is used to provide mobile real-time information about a specific traffic environment, such as congestion, accidents and events within a city. These applications are useful for transportation services, such as route planning, flow management, and fuel-efficient navigation. Traditional traffic monitoring is to analyze data captured by fixed traffic cameras, but this coverage is very low, and the data is processed by remote cloud servers, resulting in high latency and bandwidth overhead. In SSEC, the on-board sensor device collects a large amount of traffic data in real time, preprocesses the data and uploads the feature data to the edge server for further analysis. In addition, humans can use their phones to upload detailed descriptions of events.
In Figure 4, pedestrians and drivers use their respective terminal devices to jointly record event data, and the edge server sends event warnings to other drivers based on the perceived data. Transportation agencies can also check the cloud for traffic conditions within their jurisdiction to take countermeasures.
C. Crowd Abnormal Event Detection Crowd Abnormal Event Detection
The object of crowd abnormal event detection is to generate a warning of abnormal events according to the data sent by sensing equipment or people. The traditional solution of crowd abnormal event detection is based on fixed monitoring prizes and video data collected by correlation processing technology. These solutions fail when the camera is not available. Popular camera-enabled portable devices can collect geotagged images, videos, and text reported by users via apps.
As shown in Figure 5, in football matches, the sudden appearance of unidentified objects or people’s malicious behaviors may threaten the safety of players and interrupt normal matches. In the SSEC architecture, viewers can upload videos, images and text to report unusual events they find, and the cloud server then sends alerts to other viewers and emergency reports to police stations.
D. Plate Recognition
License plate recognition was originally used to track suspects by identifying their license plate numbers using a front-facing camera on a car. It changes the current process of searching for suspects, which relies too heavily on eyewitness reports, by allowing surveillance cameras to expand the k coverage area. However, using cloud servers to process large amounts of video data can result in large transfer overhead and high response latency. In SSEC, the use of road test unit can solve these problems: the vehicle locally carries out data processing to extract features, and then transmits them to the road test unit for processing. At the same time, the vehicle can also filter out some privacy information. Finally, cloud servers can send alerts to the police.
E. Crowd Video Sharing
As shown in Figure 7, crowd video sharing uses self-organizing edge devices to deliver P2P video content. Football match on a position a better audience clapped moment with poor position want to share with the audience, but using the system of the application must: 1) using the edge of the involved in computing resources to avoid the bottleneck of the system is extended 2) using network coding to reduce the video files, in order to reduce delay in network environment is poorer. SSEC can combine edge devices for computing and communication, thereby solving bandwidth problems (depending on the number and size of devices) and saving computing resources. There is a bottom-up game-theoretic decision-making process that optimizes coding and delivery to minimize delays in video delivery.
IV. RESEARCH CHALLENGES AND OPPORTUNITIES
While using SSEC brings awareness, computing, and services to the source, using edge devices for social awareness presents a list of challenges that are not fully addressed.
A. Resource Management with Rational Edge
Rational edge: Because people with edge devices are usually rational, edge devices often have inconsistent or even conflicting goals with the application. Because of the characteristics of rational edge, these two problems hinder the application of existing resource management schemes: competitive goals and asymmetric information.
Competing objectives:
On the application side, it takes minutes for edge devices to complete social awareness tasks and solve quality-of-service problems on time. Edge devices, on the other hand, care more about the cost of completing a task (energy consumption, space consumption, etc.) and will not perform a task unless there is a substantial reward. This is in stark contrast to the past when computing resources were completely collaborative and controlled by applications. A new set of computational allocations must be devised to satisfy both sides.
2) Asymmetric Information:
The application server usually has detailed information about the task, which is often closely related to the quality of service (priority and duration) that the task meets the requirements of the social aware application. On the contrary, edge devices care more about their own device state, and edge devices will not share their state information with other devices or servers, which will make the server do not have enough information to make the optimal decision of computing and dispensing.
(B) occultation Cooperativeness
Constrained cooperation: Because of selfishness, edge devices usually cooperate only when they complete their own computing tasks. However, collaboration is important in SSEC, such as task allocation to a set of edge devices for parallel and collaborative completion can reduce execution time significantly. For example, a device that does not have a camera can do nothing about the task assigned to photograph its surroundings. If the device has superior computing power, it can provide computing services to the surrounding facilities. But cooperation among edge devices is particularly challenging because: 1) rational edge devices are unwilling to cooperate unless there are sufficient rewards; 2) Various restrictions such as physical distance trust also prevent cooperation between edge devices; 3) Task dependence should be considered in cooperation.
C. contributes to Heterogeneity
Due to the heterogeneity of various edge devices, it is difficult to arrange them to cooperate to complete tasks. Since devices are owned by individuals, applications cannot pick and choose devices in a completely controlled manner. To solve this problem, some research tasks are as follows:
1) Runtime Abstraction: Runtime Abstraction
The key problem is that devices in SSEC cannot support the execution of social sensing information because of the different operating environments. Containerization technologies like Docker can abstract away some of the hardware details of a device and provide a virtual environment that provides a lightweight, portable, and high-performance sandbox to host a variety of applications. In particular, social aware application developers can “wrap” all required dependencies and the operating system itself into the Docker container of each social aware application. Such runtime abstractions allow edge devices in SSEC to provide the same interface to social aware application developers, while still providing them with “write once and run anywhere” characteristics despite the heterogeneity of SSEC devices.
2) Hardware Abstraction:
The goal of hardware resource abstraction is to make resource management easier by ignoring the details of hardware heterogeneity. The various capabilities of the device hardware in the HeteroEdge are like different workers: cCPU, GPU, and sensor worker. Each worker has the ability to describe, which is related to an estimate of the worst execution time of a task. The equipment owner can designate that worker to participate. The heterogeneous hardware abstraction follows three design rules: I) the heterogeneous edge device set should form a unified homogeneous resource pool for social sensing applications; Ii) Device owners should be able to control what resources they wish to provide to the application; Iii) Edge devices can easily track their own dynamic state and provide the necessary context information for runtime decisions and optimizations in SSEC.
3) Networking Abstraction:
Heterogeneous networks: Edge devices have various network interfaces (Bluetooth, wifi, ZigBee) and need abstract network interfaces so that programmers can deploy applications without having to worry about interfaces and details. Software-defined networks (SDN) can organize networks, services, and devices in a heterogeneous network environment. It provides apis to simplify network management, define network flow, and facilitate virtualization within the network.
However, we find that existing resource management efforts in edge computing cannot adequately handle the apparent heterogeneity in SSEC. No middleware has been developed that addresses all three levels of SSEC abstraction.
D. Robustness against Churn and Dynamic Context
Loss of equipment: Edge equipment is volatile because it kills missions to save energy, or occasionally contributes computing power but then has to stop contributing to the original mission. Also, whether a device is qualified to perform a task depends on its location. If the device moves, the task may need to be performed by another device in a suitable position. We introduce buffering to multi-phase flow applications. In this system, tasks are divided into multiple stages, in which different devices perform the calculation operations of the pipeline at each stage. If a device in the pipeline needs to exit and be replaced, it can be replaced by its neighbor in the pipeline. And pipeline design can take advantage of the fine-grained advantage of heterogeneous hardware devices, because each stage can be adapted to a particular computing platform.
Another challenge is that edge devices are very volatile and their willingness to join SSEC applications can change over time. In environmental monitoring applications, each device monitors a specific area, and when the battery status of the device changes or the location of the device changes, they no longer perform the task. If this dynamic change is not grasped, suboptimal resource allocation can make it costly for edge devices to accomplish tasks.
A: Privacy and Security
The device may expose the end user’s personal information during the mobile data phase. In license plate recognition, for example, the images taken by the device may contain incoming information, where the user lives or how he or she moves. The application needs to get a better picture of the state of each edge device in order to maximize the efficiency of task allocation by time. However, at this time, devices are reluctant to share this state information due to personal privacy. At present, privacy preservation technology, such as anonymous technology, can effectively protect device privacy. But this prevents the server from identifying the calculated contributors and distributing rewards in the traditional way.
Security is important for both the device and the application so that committed services are not corrupted. SSEC’s architecture, where data comes from edge devices and is processed on private edge devices, is not suitable for traditional security systems (authentication to access resources). 1) P2P-API interfaces of collaborative edge or hybrid SSEC architecture must be redesigned so that user information cannot be stolen; 2) Can not upload incorrect measurement information or calculation results, because can not be obtained without work; 3) The system is resistant to malicious attempts to destroy or “poison” the results of the application; 4) The system should be able to avoid denial-of-service attacks, such as maliciously delaying tasks to destroy quality of service.
It is ROADMAP FOR FUTURE WORK
A. SSEC and 5G
5G is said to be able to reach 10GB/s with a latency of less than 1ms, which will improve the SSEC status quo and facilitate the emergence of new SSEC applications. This will increase access to edge devices and facilitate collaboration. To ensure latency, 5G base stations must be deployed at high densities, which can also be used as edge servers. SSEC applications can enjoy high speed and low latency. We expect 5G and future networking technologies to support more data-intensive and delay-sensitive SSEC applications.
B. SSEC and AI
Many edge devices are low-end sensors with no processing power or hardware to support AI algorithms. SSEC can advance AI by developing the intelligent edge of collaboration, where low-end devices upload AI tasks to be performed on ai-capable devices. Set up a fair incentive mechanism to motivate devices, because electricity is very valuable for edge devices. And because the equipment is privately owned, personal and trust issues must be addressed.
C. SSEC and human-in-the-loop SSEC
Humans will become part of SSEC’s “social edge node,” where they can reason and make decisions that help perform certain critical tasks that require specialized knowledge. Building an integrated human-machine system must take full advantage of human intelligence.
VI. CONCLUSION CONCLUSION
In this paper, an emerging SSEC framework is proposed to improve the scalability and responsiveness of social sensing applications by leveraging edge-based infrastructure and a growing number of powerful iot devices. With its human-centered design, SSEC envisions integrating human intelligence into data collection, processing, analysis and decision-making processes. Several emerging applications enabled by SSEC were discussed, along with some of the open research challenges to be undertaken.
Article structure analysis
The original source
Daniel, Zhang, Vance N , et al. When Social Sensing Meets Edge Computing: Vision and Challenges[J]. 2019.
Xueshu.baidu.com/usercenter/…