Overview of edge calculation

What is edge computing

Start with Cloud Computing

Cloud computing will compute resources 、 Data resources are unified in a strong place （ Cloud Center ）, Let ordinary users become thin customers . in other words , Users store data in the cloud , The requirements for local storage capacity and computing power are reduced , Save the cost of upgrading equipment

But cloud computing also has problems , Communication has become a bottleneck , for example ：

1. Broadband is below demand ： A Boeing 787 Will be generated in every second 5G data , But bandwidth （ Whether it's a ground base station 、 satellite ） Far from keeping up with this requirement
2. Data communication takes time （ The real time ）： for example , Autopilot , Every second produces 1G data , If this 1G The data is transferred to the cloud to complete the calculation , The communication time required is 800ms（10Gbps by 5G), If the vehicle speed is 60km/h,800ms It's past 13 rice
3. High energy consumption ：ZigBee The calculated energy consumption unit is uW, The unit of energy consumption for data transmission is mW, Compared with the energy consumed by data calculation , The transmission energy consumption of the cloud is too large
4. Security ： Worried that putting photos in the cloud will reveal , Privacy and confidentiality issues

Therefore, the concept of edge calculation is proposed

Edge of computing ： Edge computing refers to processing with network edge nodes 、 Analyze the data . among , Edge node refers to any node with computing resources and network resources between the data generation source and the cloud center .** To be frank ： That is to place the calculation near the data source .** The gray wave at the bottom left in the figure below is the edge node used for calculation , Compared to cloud computing , They are close to the local area , The problem of cloud computing is partially solved .

The problem to be solved by edge computing

1. Solve high reliability 、 The need for low latency computing power , This demand is often needed in industry
2. Solve the problem of excessive energy consumption in the data center
3. Solve the pressure of big data processing ： The data generated by the Internet of things will grow exponentially , The network cannot deliver , The data center cannot handle
4. Solve security problems ： Private data is not uploaded to the data center , Reduce security risks

The case of edge computing

1. The shopping cart
   Every operation is synchronized to the cloud （ Packet loss of operation data may occur , At this time, local data and cloud data are no longer synchronized , There's chaos ） or Synchronize to edge nodes , After a period of time, the information of edge nodes will be synchronized to the cloud at one time （ The possibility of packet loss is greatly reduced , There is something similar to the checkpoint of database ）
   
2. traffic
   Internet cars need real-time , Calculate after transferring data to the cloud , Then send the calculation result back , It is not feasible to judge whether someone in front needs to turn , Data must be calculated locally to ensure real-time , here , A computing device in a car can be used as an edge node
3. Medical care
   The average hospital bed is 20 More than networked devices , A lot of data will be generated . Patient data （ privacy ） Processing takes place closer to the edge , Instead of sending it to the cloud , It can avoid the risk of improper access to data .

The relevant formulation of edge calculation

1. Fog calculation ：2011 By Cisco (Cisco) Put forward .
   
   In fact, there is no essential difference between fog calculation and edge calculation , All of them place the calculation near the data source
2. Mobile edge computing ： European Telecommunications Standardization Association proposed mobile edge computing （Mobile Edge Computing,MEC）
   The key concept is the sinking of some mobile services , Improve its business distribution 、 Transmission capacity , Further reduce delay , And effectively suppress the congestion in the core network
   

Edge calculation reference frame

up to now , There is no good framework for edge computing , The frame is for reference only

Edge calculation reference frame – manufacture

Edge computing core technology challenges

Programming (Program)

Windows In the environment exe Can't be in Linux Up operation , Different operating systems run in different environments ,x86 and ARM stay CPU The instruction set is different , For edge computing , Heterogeneous is very common , Such as raspberry pie based on ARM CPU、NVIDIA Jetson Nano be based on GPU、iPhone be based on iOS、Huawei Mate 20 be based on Android wait …
Same function , The requirements for running in different environments are different , If you rewrite every isomer , The price is too high. .
Core challenges ： How to write an application that can be distributed and deployed in the edge computing framework

name (Naming)

name (Naming), That is, locate resources . for example , The monitoring system requires a certain camera , If it is a stand-alone operation , You can name the camera according to the file path , for example ,open(/dev/camera0). But the edge computing framework may use non local cameras , How to deal with ？ A simple idea is through IP Address to find the specific node ,IPV4 Is it enough ？IPV6 Is there a specific and clear meaning ？ If there are many cameras in many different places , We have to write down a bunch of things that have no rules IP The address? ？
Core challenges ： How to name resources to make them suitable for large-scale dynamic heterogeneous environments ?

Isolation

Computing resource isolation ： That is, applications cannot interfere with each other .
Data isolation ： That is, different applications should have different access rights .
Edge calculation is more complex , It's easier to make mistakes , But it requires more reliability
Docker: Commonly used Docker Technology can be applied on the basis of 0S Level virtualized isolated environment
docker Introduce ：https://www.cnblogs.com/idktp/articles/10538872.html

Dynamic environment

Just like the Internet , Edge computing also forms a network , It's dynamic , New nodes may be added at the next moment , There will also be the exit of old nodes . Dynamic environment has three steps ： Service discovery 、 Fast configuration and load balancing .
Service discovery ： Calculate how the service requester knows the surrounding services .
Quick configuration ： Mobile node 、 Device switching will cause dynamic registration and revocation of services , Services usually need to be migrated along with it ……
Load balancing ： data 、 demand 、 service 、 The Internet 、 Calculation 、 Many factors such as energy consumption determine load balancing

Edge operating system

• Robot operating system (robot operating system, ROS) It is a set of open source robot development and management tools , Provide hardware abstraction and drivers 、 Message communication standard 、 Software package management, etc A series of tools , It is widely used in industrial robots 、 Edge computing scenarios such as autonomous vehicles, unmanned aerial vehicles .
• EdgeOSH It is an edge operating system designed for smart home , It is deployed in the edge gateway of the home , adopt 3 Layer function abstractly connects the upper application and the lower smart home hardware .
• EdgeOSv It is an edge operating system for Internet connected vehicles

ROS It may become a typical operating system for edge computing scenarios

Algorithm execution framework

• AI computing requires an execution framework , In fact, any calculation has an execution framework , Such as desktop computing
• On the data center TensorFlow： Perform model training tasks , The input is a large batch data set , Pay attention to the iteration speed during training 、 Convergence rate and scalability of the framework
• On the edge node TensorFlow： Perform forecast task , The input is real-time small-scale data , Pay attention to the prediction speed 、 Memory usage and energy efficiency

Typical framework of edge computing

OpenVDAP（2018, Wayne State University ）
KubeEdge（ Huawei ）

• Edged： management Edge Container application on .
• EdgeHub：Edge Communication interface module .
• CloudHub： Communication interface module of cloud .
• EventBus： Use MQTT Processing internal edge communication .
• DeviceTwin： It is the software image of the device that processes the metadata of the device
• MQTT： Message Queuing Telemetry Transport, Is a subscription based / released IoT Connection protocol

Azure IoT Edge（2017, Microsoft ）
OpenEdge（2018.12, Baidu ）

Harbin Institute of technology 《 Edge computing and embedded intelligence 》 note