什么是边缘计算边缘计算_什么是边缘计算

什么是边缘计算边缘计算

The concept of Edge Computing is inspired by CDN technology. CDN stands for Content Delivery Networks. A CDN typically works to bring the content (images, video, script files) on the Internet closer to its users. This helps faster streaming of content with proper load handling. This is how YouTube, Netflix, etc delivery content to different regions without getting overwhelmed by the massive data rates required for streaming services.

边缘计算的概念受到CDN技术的启发。 CDN代表内容交付网络。 CDN通常用于使Internet上的内容(图像，视频，脚本文件)更接近其用户。 这有助于通过适当的负载处理来更快地传输内容。 这样，YouTube，Netflix等将内容传送到不同的区域而不会被流式传输服务所需的海量数据速率所淹没。

Edge Computing brings both data and computations closer to its users!

边缘计算使数据和计算更接近其用户！

In Edge Computing the data and computation are localized so that the response times and bandwidth requirements are significantly reduced. This also supports Green Computing, where the computations are done with minimal use of resources.

在Edge Computing中，数据和计算被本地化，因此响应时间和带宽需求大大减少。 这也支持“绿色计算” ，即在使用最少资源的情况下完成计算。

边缘计算的起源 (Origin of Edge Computing)

With the increase of network-attached devices due to the popularity of IoT, the amount of data generated has increased exponentially. This increase demands massive computing and analytics at data centres which increase the utilization of bandwidth.

随着物联网的普及，随着网络连接设备的增加，生成的数据量呈指数增长。 这种增加要求在数据中心进行大规模的计算和分析，从而增加带宽利用率。

In Edge Computing the computational component is pushed towards the ends of the network, i.e. the users’ end. This eliminates the need for servers to continuously work on behalf of each user, but rather invest their time on analytics at a much higher level (e.g. community analytics, community-based recommendations, etc). So how would personal data be processed?

在边缘计算中，计算组件被推向网络的末端，即用户的末端。 这样就不再需要服务器代表每个用户连续工作，而是将他们的时间投入到更高层次的分析上(例如，社区分析，基于社区的建议等)。 那么如何处理个人数据？

Data is processed at the users' end using the connected devices such as smart phones, smart home hubs, smart TVs, etc.

使用连接的设备(如智能手机，智能家居集线器，智能电视等)在用户端处理数据。

For example, your sleep data is processed within your phone, where most data for this process is collected either from your phone, watch or using all devices with the same account logged in. Also, iPhone face detection runs completely offline and learns continuously based on your facial changes that take place over time.

例如，您的睡眠数据是在手机中处理的，该过程的大多数数据是从手机，手表或使用登录了相同帐户的所有设备收集的。此外，iPhone面部检测功能完全脱机运行，并根据随时间变化的面部变化。

Photo by Szabo Viktor on Unsplash

Szabo Viktor在 Unsplash上 拍摄的照片

Let’s consider a hypothetical example (is it?) and explore how Edge Computing and Cloud Computing would approach this problem

让我们考虑一个假设的示例(是吗？)，并探讨边缘计算和云计算将如何解决此问题。

There is an organization, and they design beauty facial equipment such as sun-glasses, eyeliners, lipstick, etc. They want to gather data from 1000 voluntary participants. This happens by means of an app, that lets them take a picture of their own followed by few edit actions to dress with makeup or shades. The app also allows them to select their face through a square overlay.

有一个组织，他们设计美容仪器，例如太阳镜，眼线笔，口红等。他们希望从1000名自愿参加者那里收集数据。 这是通过一个应用程序实现的，该应用程序使他们可以自己拍摄照片，然后进行一些编辑操作以化妆或遮盖阴影。 该应用程序还允许他们通过正方形叠加选择自己的脸。

Note that this is a simple example where the data is originated at users’ end. There can be more complex scenarios where data is shared between users and analysed locally within houses (home users), etc.

请注意，这是一个简单的示例，其中数据起源于用户端。 可能存在更复杂的场景，其中用户之间共享数据并在房屋(家庭用户)等本地进行分析。

云计算方法 (Cloud Computing Approach)

The data will be gathered as images and makeup effect coordinates. These will be uploaded into the cloud storages. These data will then be read again followed by classification based on pre-defined metrics to classify images based on skin-tone, hair-colour, etc. This approach has the following concerns.

数据将作为图像和化妆效果坐标收集。 这些将被上传到云存储中。 然后将再次读取这些数据，然后再基于预定义的指标进行分类，以基于肤色，头发颜色等对图像进行分类。此方法存在以下问题。

• Data is uploaded which is a bit of a privacy concern
  数据已上传，这有点隐私问题
• Requires bandwidth and cloud storage data for thousands of images+other annotations (makeup, etc)
  需要用于数千张图像的带宽和云存储数据以及其他注释(化妆等)
• Cloud computers need to have sufficient memory and computing power to handle images
  云计算机需要具有足够的内存和计算能力来处理图像
• Requires time to gather data and a re-run of the analytics pipeline could take a lot of time
  需要时间来收集数据，并且重新运行分析管道可能会花费大量时间

Photo by Mateus Campos Felipe on Unsplash

Mateus Campos Felipe在 Unsplash上的 照片

边缘计算方法(Edge Computing Approach)

The data is analysed at users’ end for most of it. For example, the following could be done at the users’ end.

大部分数据都在用户端进行分析。 例如，可以在用户端进行以下操作。

• Facial recognition which is assisted by the user him/her self by annotating face area unknowingly
  使用者在不知情的情况下通过注释自己的面部区域来辅助自己的面部识别
• Detection of skin-tone, hair colour and fitting a face model that converts the entire face into a numeric matrix/coordinate encoding
  检测肤色，头发颜色并拟合将整个脸部转换为数字矩阵/坐标编码的脸部模型
• Makeup information is converted into numerical annotations with intensity, opacity alongside with other matrics
  化妆信息与其他矩阵一起转换为具有强度，不透明度的数字注释
• The data required for analysis is sent over to the server in a simple JSON format
  分析所需的数据以简单的JSON格式发送到服务器
• The analysis in the cloud is now reduced. The data is transferred as text rather than blobs (images)
  现在减少了云中的分析。 数据以文本而不是斑点(图像)的形式传输

As we can see the weight on the cloud has been transferred towards the edge of the network, in this case, the smartphones. Users are more than willing to take part in such mobile apps, which of course leave them with gorgeous captures of themselves. Sounds familiar?

我们可以看到，云上的权重已经转移到了网络边缘，在这种情况下是智能手机。 用户非常愿意参与这样的移动应用程序，这当然使他们拥有自己的精彩捕获。 听起来很熟悉？

Photo by Ildefonso Polo on Unsplash

Ildefonso Polo在 Unsplash上 拍摄的照片

边缘计算的含义(Implications of Edge Computing)

Edge Computing demands different measures of privacy, with unimaginable scalability and potential for future data analytics demands. The distribution of computation over to the edge of the network thus makes a new paradigm of computation.

边缘计算要求采取各种不同的隐私措施，并具有难以想象的可扩展性和未来数据分析需求的潜力。 因此，计算在网络边缘的分布成为了一种新的计算范式。

What changes? and how do these aspects change? What’s there for us?

有什么变化？ 这些方面如何变化？ 我们有什么呢？

隐私权与安全性 (Privacy and Security)

Data needs to be securely transmitted no matter what their format is. This has already been an established aspect of networking and computing. Encryption schemes are required for this task. Sometimes data can be shared between users and in such scenarios encryption is needed. There are encryption schemes that enable operations on encrypted data that can be decrypted later for viewing (ref). This is known as holomorphic encryption.

无论数据格式如何，都需要安全地传输数据。 这已经是网络和计算的既定方面。 此任务需要加密方案。 有时可以在用户之间共享数据，并且在这种情况下需要加密。 有一些加密方案可以对加密数据进行操作，这些数据以后可以解密以供查看( ref )。 这称为全同加密。

可扩展性 (Scalability)

Edge computing provides immense scalability. However, the algorithms and other computations must be carefully broken so that the end result is nearly equal, had it been performed in a cloud. This all should happen to wit the idea of heterogeneity in mind. It is always challenging to work in distributed environments. But the efforts are most rewarding. The example we discussed impose an embarrassingly parallel computation scenario if we think of facial recognition.

边缘计算提供了巨大的可扩展性。 但是，必须谨慎地打破算法和其他计算，以使最终结果与在云中执行的结果几乎相等。 考虑到异质性的想法，所有这些都会发生。 在分布式环境中工作始终是挑战。 但是这些努力是最有意义的。 如果我们想到面部识别，我们讨论的示例将产生一个令人尴尬的并行计算场景。

可靠性 (Reliability)

Although the scalability of massive, the failures are inevitable in a distributed environment. This is because in this scenario the edge users are not committed and can leave the system without a graceful disconnection. Hence, backup mechanisms must be accompanied for better reliability. It is also crucial that relevant analytics are done with resource constraints in mind. Otherwise, device failures can cause without leaving a trace.

尽管规模可扩展，但在分布式环境中不可避免地会发生故障。 这是因为在这种情况下，边缘用户没有提交，并且可以在没有正常断开连接的情况下离开系统。 因此，必须附带备份机制以提高可靠性。 同样重要的是，在进行相关分析时要牢记资源限制。 否则，可能会导致设备故障而无法追踪。

速度 (Speed)

The speeds of analytics can be improved since the computations are done locally. This also could improve results due to lower bandwidth requirements to fetch data. In case of a recommendation system that runs within your phone’s shopping app, the app itself can fetch your likely shopping items without having to make multiple API calls.

由于计算是在本地完成的，因此可以提高分析的速度。 由于较低的带宽要求以获取数据，因此这也可以改善结果。 如果推荐系统在手机的购物应用程序中运行，则该应用程序本身可以获取您可能的购物商品，而无需进行多个API调用。

效率 (Efficiency)

Usually, Edge Computing tries to bring computations closer to the users who will likely use the results. Hence, the systems as such as obviously efficient given that they do not use many other resources.

通常，Edge Computing尝试使计算更接近可能会使用结果的用户。 因此，这样的系统在不使用许多其他资源的情况下显然很有效。

Edge的现有库和设备 (Existing Libraries and Devices for Edge)

It would be pointless if I did not provide a starting point for readers to get engaged with Edge Computing. Luckily there are many libraries that support analytics in resource-constrained environments. However, the development of a sophisticated system could yet be challenging.

如果我没有为读者提供参与Edge Computing的起点，那将毫无意义。 幸运的是，有许多库在资源受限的环境中支持分析。 但是，开发复杂的系统可能仍然具有挑战性。

边缘设备的ML框架 (ML Frameworks for Edge Devices)

• ML Kit and TensorFlow-Lite by Google

  Google提供的ML Kit和TensorFlow-Lite

• Firebase ML Kit

  Firebase ML套件

• PyTorch mobile

  PyTorch手机

• Core ML by Apple

  苹果的核心ML

支持的设备 (Supported Devices)

• Raspberry Pi 4/3/2/1/0
  树莓派4/3/2/1/0
• Compute modules by NVidia, Intel, etc
  NVidia，Intel等计算模块
• Android devices
  Android设备
• Home Cloud-based NAS devices (Android/Linux based devices that run on arm chips/SOCs)
  基于家庭云的NAS设备(在手臂芯片/ SOC上运行的基于Android / Linux的设备)

I hope you enjoyed reading this article. Cheers!

希望您喜欢阅读本文。 干杯!

Here’s another article that you might like reading;

这是您可能想阅读的另一篇文章。

翻译自: https://medium.com/swlh/what-is-edge-computing-d27d15f843e

什么是边缘计算边缘计算