This article is reproduced from：https://blog.csdn.net/gui951753/article/details/80952907

This translation of Professor Shi Weisong's paper《Edge Computing : Vision and Challenges》

Abstract

With the rapid development of Internet of Things technology and the promotion of cloud services, cloud computing model can no longer solve the current problem well. Therefore, a new computing model, edge computing, is given here. Edge computing refers to processing data at the edge of the network, which reduces request response time, improves battery life, reduces network bandwidth, and guarantees data security and privacy. This article introduces the concepts of edge computing through a number of cases, including cloud offloading, smart home, smart city, and collaborative edge nodes for edge computing. I hope this article will give you some inspiration and get more people involved in edge computing research.

Brief

Cloud computing has gradually changed the way we live, learn and work since it was introduced in 2005. The services provided by software such as Google and facebook, which are often used in life, are typical representatives. And scalable infrastructure and processing engines that support cloud services also have an impact on the pattern of our carrier industry, such as hadoop, spark, and so on.

The rapid development of the Internet of Things has brought us into the post-cloud era, which produces a large amount of data in our daily life. Cisco estimates that nearly 50 billion things will be connected to the Internet by 2019. Internet of Things applications may require extremely fast response times, data privacy, and so on. If the data transferred from the Internet of Things to the cloud computing center, it will increase the network load, the network may cause congestion, and there will be some data processing delay.

With the advent of the Internet of Things and cloud services, we have assumed a new model for solving problems, edge computing. Generate, process, and analyze data at the edges of the network. The next article will explain why edge computing is needed and the definitions. Some research on cloud offloading and smart cities, including programming, naming, data abstraction, service management, data privacy, and security under edge computing, will also be discussed below.

What is edge computing

The data generated at the edge of the network is increasing gradually. This computing model will be more efficient if we can process and analyze the data at the edge nodes of the network. Many new computing models are being proposed because we find that with the development of the Internet of Things, cloud computing is not always so efficient. Edge computing refers to processing and analyzing data at network edge nodes. Here, we define an edge node, which refers to any node with computing and network resources between the source of data generation and the center of the cloud. For example, mobile phones are the edge nodes between people and cloud centers, and gateways are the edge nodes between smart home and cloud centers. In an ideal environment, edge computing refers to the analysis and processing of data near the source of data generation, without data flow, thereby reducing network traffic and response time. The next article will list some reasons why edge computing is more efficient and better than cloud computing.

Why Edge Calculation is Needed

Driven by Cloud Services: Cloud Centers have powerful processing capabilities and can handle large amounts of data. However, transferring large amounts of data to the center of the cloud is a challenge. The system performance bottleneck of the cloud computing model is the limitation of network bandwidth, it takes a certain time to transfer a large amount of data, and it also takes a certain time for the cloud center to process data, which will increase the request response time and make the user experience very poor.

The driving force of the Internet of Things: Almost all electronic devices can now connect to the Internet, and these electronic devices will generate a large amount of data later. The traditional cloud computing model can not process these data in time and effectively. Processing these data at edge nodes will result in minimal response time, reduce network load, and ensure the privacy of user data.

Role Change of Terminal Devices: Terminal devices spend most of their time playing the role of data consumers, such as using smartphones to watch iQIYI, brush jitter, and so on. However, now smartphones give terminal devices the ability to produce data, such as purchasing things in Taobao and searching for content in Baidu, which are data generated by terminal nodes.

Below are two diagrams. Figure 1 is a paradigm under the traditional cloud computing model. On the far left is a service provider that provides data, uploads it to the cloud center, sends a request to the cloud center, and the cloud center responds to the request and sends the data to the end customer. End customers are always the role of consumers.

Figure 2 is an edge computing paradigm with the rapid development of the Internet of Things. Edge nodes (including smart appliances, mobile phones, tablets, etc.) generate data that is uploaded to the cloud center, and service providers generate data that is uploaded to the cloud center. Edge nodes send requests to the cloud center, which returns related data to edge nodes.

 

 

Advantages of edge computing

In face recognition, the response time is reduced from 900ms to 169 ms, and the overall system's energy consumption is reduced by 30%-40% after some computing tasks are unloaded from the cloud to the edge. Data integration, migration, etc. can be reduced by 20 times.

Case Study

Cloud Uninstall

In traditional content distribution networks, data is cached to edge nodes. With the development of the Internet of Things, data production and consumption are at edge nodes, that is, edge nodes need to undertake certain computing tasks. The process of offloading computing tasks from the center of the cloud to edge nodes is called cloud offloading.

For example, the development of the mobile Internet has enabled us to shop smoothly on the mobile side. Our shopping carts and related operations (the addition or deletion of goods) depend on uploading data to the cloud center. If the related data and operations of the shopping cart are all lowered to the edge nodes, it will greatly improve the response speed and enhance the user experience. Improve the quality of human-system interaction by reducing latency.

Video Analysis

With the increase of mobile devices and the increase of camera placement in cities, using video to achieve a certain purpose has become an appropriate means, but cloud computing is no longer a suitable model for this kind of video processing, because the transmission of large amounts of data over the network may lead to network congestion, and the privacy of video data cannot be guaranteed.

Therefore, edge computing allows the cloud center to place related requests. Each edge node processes the requests in combination with local video data and then returns only the relevant results to the cloud center. This reduces network traffic and guarantees user privacy to some extent.

For example, if a child is lost in a city, the cloud center can place a request to find the child at each edge node, which is processed in conjunction with local data and then returns the result whether the child was found or not. This is a faster way to solve problems than uploading all your videos to the cloud center and letting the cloud center solve them.

Smart home

The development of the Internet of Things has made electronic devices in ordinary people's homes more active. Just connecting these electronic devices to the network is not enough. We need to make better use of the data generated by these electronic components and make better use of the data to serve the current home. Given network bandwidth and data privacy protection, we need data that can best only be circulated locally and processed locally. We need a gateway as an edge node so that it consumes the data generated in the home itself. At the same time, because there are many sources of data (can be from any smart device, such as computers, mobile phones, sensors), we need to customize a special OS so that it can knead these abstract data together and integrate them organically.

Smart city

Edge computing is designed to bring data closer to the data source, so edge computing has several advantages in smart cities:

Mass data processing: In a large city with a large population, a large amount of data is generated all the time, and if the traffic is handled by the cloud center, it will result in a huge network burden and a serious waste of resources. If the data can be processed close to each other and within the local area network where the data source is located, the network load will be greatly reduced and the processing power of the data will be further improved.

Low latency: In large cities, there are many services that require real-time characteristics, which require that the response speed be as fast as possible. For example, in medical and public security, edge computing reduces the time data travels across the network, simplifies the network structure, and allows the analysis, diagnosis and decision-making of data to be handled by edge nodes to improve the user experience.

Location awareness: For some location-based applications, the performance of edge computing is due to cloud computing. For example, in navigation, terminal devices can give the related location information and data to edge nodes based on their real-time location for processing, and edge nodes can make judgments and decisions based on existing data. The network overhead is minimal throughout the process. User requests were responded very quickly.

Smart Car

Cloud computing is indispensable to the development of smart networked automotive and automotive driving technology, and edge computing is more practical in this field. The high delay and poor stability in 4G network environment limit the development of auto-driving technology for smart cars. In 5G environment, the implementation of auto-driving technology has the basic conditions. Edge computing can sink more data computing and storage from the "core" to the "edge" and deploy it close to the data source. Some data does not have to go through the network to reach the cloud anymore, such as high-precision maps and real-time image recognition. Edge computing can reduce latency and network load, and also improve data security and privacy. This is self-evident in the auto-driving field where time delay is extremely demanding and data processing and storage is enormous. In the future, mobile communication devices close to vehicles, such as base stations, roadside units, and so on, will deploy edge computing for vehicle networking to complete local-end data processing, encryption and decision-making, and provide real-time, highly reliable communication capabilities.

Edge Collaboration

Some data cannot be processed by cloud centers due to data privacy issues and the cost of data transmission over the network, but sometimes it requires collaboration between multiple departments to make the most of it. Thus, we propose the concept of edge collaboration, which uses multiple edge nodes to collaborate, create a virtual view of shared data, and integrate these data using a predefined public service interface through which we can write applications to provide more complex services to users.

An example of how multiple edge nodes work together to win. For example, when an influenza outbreak occurs, hospitals as an edge node share data with multiple nodes such as pharmacies, pharmaceutical companies, governments, insurance industries, and so on, sharing the number of people infected with the current influenza, the symptoms of influenza, the cost of influenza treatment, etc. to the above edge nodes. With this information, the pharmacy can adjust its purchasing plan and balance the warehouse inventory. Pharmaceutical companies can prioritize the production of medicines that matter by sharing data. The government has raised the level of influenza alert to people in relevant areas, and further action can be taken to control the spread of influenza outbreaks. Insurance companies will adjust the price of this type of insurance next year based on the severity of the flu. In summary, any of the edge nodes will benefit from this data sharing.

Opportunities and Challenges

These are the potential and prospects of edge computing in solving related problems. Next, we will analyze the opportunities and challenges that will be faced in the process of implementing edge computing.

Programming Feasibility

Programming on a cloud computing platform is very convenient because the cloud has a specific compilation platform, and most programs can run on the cloud. But programming under edge computing will face a problem, platform heterogeneity, each network edge is different, it may be IOS system, Android or linux, and so on, programming under different platforms is different. Therefore, we propose the concept of computational flow, which is a function sequence/computational sequence along a data propagation path, through which applications can specify which node to compute in the data propagation path. Computing streams can help users determine what functions/calculations should be completed and how data is propagated after calculations occur at the edge. By deploying a compute stream, you can keep your calculations as close to the data source as possible.

Name

Naming schemes are important for programming, addressing, object identification, and data communication, but there is no effective way to process data in edge computing. Things in edge computing have a variety of communication technologies, such as wifi, bluetooth, 2.4g, etc. Therefore, relying solely on the tcp/ip protocol stack cannot satisfy the communication between these heterogeneous things. Naming schemes for edge computing need to deal with the mobility of things, dynamic network topology, privacy and security protection, and scalability of things. Traditional naming mechanisms, such as DNS (Domain Name Resolution Service) and URI (Uniform Resource Marker), cannot solve the naming problem of dynamic edge networks very well. The NDN (Named Distribution Network) currently being proposed to solve these problems also has some limitations. In a relatively small network environment, we propose a solution. As shown in Figure 3, we describe when, where, and what a thing is doing. This unified naming mechanism makes management very easy. Of course, when the environment rises to the height of a city, this naming mechanism may not be appropriate and can be discussed further.

Data Abstraction

There will be a large amount of data generation in the Internet of Things environment, and due to the heterogeneous environment of the Internet of Things network, the generated data is in a variety of formats. Formatting a variety of data is a challenge for edge computing. At the same time, most things on the edge of the network only collect data periodically, sending the collected data to the gateway periodically, but the storage in the gateway is limited, he can only store the latest data, so the data of edge nodes will be refreshed frequently. Using an integrated data table to store data of interest, the internal structure of the table can be shown in Figure 4, where the data is represented by id, time, name, data, and so on.

If too much original data is filtered out, the report of edge node data will be unreliable. If a large amount of original data is retained, the storage of edge nodes will be a new problem. At the same time, these data should be read, write and operate by referencing programs. Because of the heterogeneity of things in the Internet of Things, there will be some problems in reading, writing and operation of databases.

Service Management

Service management of edge nodes We think there should be four features, including differentiation, scalability, isolation and reliability, to ensure an efficient and reliable system.

Differentiation: With the development of the Internet of Things, there will be such a variety of services, different services should have different priority. For example, critical services such as things judgment and malfunction alerts should be higher than other general services, and services related to human health such as heart rate detection should have a higher priority than entertainment-related services.

Scalability: Items in the Internet of Things are dynamic. It is not easy to add or delete an item to the Internet of Things. The lack of services or the ability to add a new node to adapt to them are all issues to be solved, which can be solved by a highly scalable and flexible design of edge os.

Isolation: Isolation means that different operations do not interfere with each other. For example, there are multiple applications that can control the light in your home. Data about controlling the light is shared, and when one application cannot respond, other applications can still control the light. That is, these applications are independent of each other and have no influence on each other. Isolation also requires that user data and third-party applications be isolated, that is, applications should not be able to track and record user data. To solve this problem, a new way for applications to access user data should be added.

Reliability: Reliability can be discussed in terms of services, systems, and data

On the service side, the loss of any node in the network topology may result in the unavailability of the service, which can be avoided if the edge system can detect high-risk nodes in advance. A better way to do this is to use wireless sensor networks to monitor server clusters in real time.

From a system perspective, the edge operating system is an important part of maintaining the entire network topology. Status and diagnostic information can be exchanged between nodes. This feature makes it very convenient to deploy fault detection, node replacement, data detection at the system level.

From a data perspective, reliability means that data is reliable in sensing and communication. Nodes in an edge network may report information when they are unreliable, such as when the sensor is low on power, which most likely results in unreliable data being transmitted. To solve these problems, new protocols may be proposed to ensure the reliability of the Internet of Things in data transmission.

Privacy

The existing way to provide services is to upload data from end users of mobile phones to the cloud, then use the powerful processing power of the cloud to process tasks. During data uploading, data can easily be collected by other interested people. To keep the data private, we can start with the following aspects.

1. Processing user data at the edge of the network so that data is only stored, analyzed, and processed locally.

2. Restrict access to private data by setting permissions for different applications.

3. Edge networks are highly dynamic networks and require effective tools to protect the transmission of data across the network.

Optimization Index

In edge computing, because there are many nodes and the processing capacity of different nodes is different, it is important to select appropriate scheduling strategies among different nodes. Next, the optimization criteria are discussed in terms of latency, bandwidth, energy consumption and cost.

Delay: It is clear that cloud centers have strong processing power, but network latency is not only determined by processing power, it also combines the time data travels across the network. Take smart city distance for example, if you are looking for lost child information, processing it locally on your mobile phone and returning the results to the cloud can significantly speed up the response. Of course, this kind of thing is also relative. We need to put a logical judgment layer to decide which node is the right one to handle the task. If the mobile phone is playing games or handling other very important things at this time, the mobile phone is not very suitable for this kind of task. It would be better to leave this task to another level.

Bandwidth: High bandwidth transfers data with low latency, but high bandwidth also means a lot of waste of resources. There are two possibilities for data processing at the edge. One is that the data is completely processed at the edge and then uploaded to the cloud at the edge node. Another result is that part of the data is processed, and the rest is handed over to the cloud. Either of these two ways can greatly improve the status of network bandwidth, reduce data transmission across the network, and enhance the user experience.

Energy consumption: For a given task, you need to decide whether to save resources by placing it locally or transferring it to other nodes for calculation. If local is idle, local computing is of course the most resource-efficient, and if it is busy locally, it is more appropriate to assign computing tasks to other nodes. It is very important to balance the energy consumed by calculation and network transmission. Generally, when network transmission consumes much less resources than local computing consumes, we consider using edge computing to offload computing tasks to other idle nodes to help achieve load balancing and ensure high performance for each node.

Cost: Current costs in edge computing include, but are not limited to, the construction and maintenance of edge nodes, the development of new models, and so on. Using an edge computing model, large service providers can make more profit while doing the same work.

Summary

With the development of the Internet of Things and the promotion of cloud computing, edge computing models are emerging in the community. Processing data at edge nodes can improve response speed, reduce bandwidth, and ensure privacy of user data. In this article, we propose that edge computing may be used in some related scenarios in future life, and also mention the prospects and challenges for the future development of edge computing. It is believed that in the background of 5G and the interconnection of all things, edge computing will attract more attention and produce a wider range of applications.