自动驾驶发展_自动驾驶网络及其发展
自動駕駛發展
介紹 (Introduction)
Talking about inspiration in the networking industry, nothing more than Autonomous Driving Network (ADN). You may hear about this and wondering what this is about, and does it have anything to do with autonomous driving vehicles? Your guess is right; the ADN concept is derived from or inspired by the rapid development of the autonomous driving car in recent years.
談到網絡行業的靈感，無非就是自動駕駛網絡(ADN)。 您可能聽說過這，并想知道這是什么，它與自動駕駛汽車有什么關系嗎？ 您的猜測是正確的； ADN概念源于自動駕駛汽車近年來的快速發展或受其啟發。
Driverless Car of the Future, the advertisement for “America’s Electric Light and Power Companies,” Saturday Evening Post, the 1950s. Credit: The Everett Collection. (Mark W., 2014)未來無人駕駛汽車，“美國的電燈和電力公司”的廣告，1950年代，星期六晚上郵報。 信用：Everett收藏。 (馬克·W，2014年)The vision of autonomous driving has been around for more than 70 years. But engineers continuously make attempts to achieve the idea without too much success. The concept stayed as a fiction for a long time. In 2004, the US Defense Advanced Research Projects Administration (DARPA) organized the Grand Challenge for autonomous vehicles for teams to compete for the grand prize of $1 million. I remembered watching TV and saw those competing vehicles, behaved like driven by drunk man, had a really tough time to drive by itself. I thought that autonomous driving vision would still have a long way to go. To my surprise, the next year, 2005, Stanford University’s vehicles autonomously drove 131 miles in California’s Mojave desert without a scratch and took the $1 million Grand Challenge prize. How was that possible? Later I learned that the secret ingredient to make this possible was using the latest ML (Machine Learning) enabled AI (Artificial Intelligent ) technology.
自動駕駛的愿景已經存在了70多年了。 但是工程師們不斷地嘗試實現這一想法，但并沒有取得太大的成功。 這個概念長期以來一直是虛構的。 2004年，美國國防高級研究計劃局(DARPA)組織了無人駕駛汽車大挑戰，各車隊爭奪100萬美元的大獎。 我記得看電視，看到那些競爭的車輛，表現得像醉漢一樣，很難開車。 我認為自動駕駛的視野還很長。 令我驚訝的是，第二年，2005年，斯坦福大學的汽車在加州莫哈韋沙漠無人駕駛自動行駛131英里，并獲得了100萬美元的“挑戰大獎”。 那怎么可能？ 后來我得知，實現這一目標的秘密因素是使用最新的ML(機器學習)支持的AI(人工智能)技術。
Since then, AI technologies advanced rapidly and been implemented in all verticals. Around the 2016 time frame, the concept of Autonomous Driving Network started to emerge by combining AI and network to achieve network operational autonomy. The automation concept is nothing new in the networking industry; network operations are continually being automated here and there. But this time, ADN is beyond automating mundane tasks; it reaches a whole new level. With the help of AI technologies and other critical ingredients advancement like SDN (Software Defined Network), autonomous networking has a great chance from a vision to future reality.
從那時起，人工智能技術飛速發展，并在各個領域得到了實現。 在2016年左右的時間里，自動駕駛網絡的概念開始出現，它通過將AI和網絡相結合來實現網絡運營自主性。 自動化概念在網絡行業中并不是什么新鮮事物。 網絡操作在這里和那里不斷地自動化。 但是這次，ADN超出了自動執行日常任務的范圍； 它達到了一個全新的水平。 借助AI技術和SDN(軟件定義網絡)等其他關鍵要素的進步，從愿景到未來現實，自主網絡都是一個巨大的機會。
In this article, we will examine some critical components of the ADN, current landscape, and factors that are important for ADN to be a success.
在本文中，我們將研究ADN的一些關鍵組成部分，當前情況以及對ADN成功至關重要的因素。
愿景 (The Vision)
At the current stage, there are different terminologies to describe ADN vision by various organizations.
在當前階段，有各種術語來描述各個組織的ADN愿景。
Even though slightly different terminologies, the industry is moving towards some common terms and consensus called autonomous networks, e.g. TMF, ETSI, ITU-T, GSMA. The core vision includes business and network aspects. The autonomous network delivers the “hyper-loop” from business requirements all the way to network and device layers.
盡管術語略有不同，但業界仍在朝著一些稱為自治網絡的通用術語和共識邁進，例如TMF，ETSI，ITU-T，GSMA。 核心愿景包括業務和網絡方面。 自治網絡從業務需求一直到網絡和設備層都提供“超環”。
On the network layer, it contains the below critical aspects:
在網絡層，它包含以下關鍵方面：
Intent-Driven: Understand the operator’s business intent and automatically translate it into necessary network operations. The operation can be a one-time operation like disconnect a connection service or continuous operations like maintaining a specified SLA (Service Level Agreement) at the all-time.
目的驅動：了解運營商的業務意圖并將其自動轉換為必要的網絡操作。 該操作可以是一次性操作，例如斷開連接服務，也可以是連續操作，例如始終保持指定的SLA(服務水平協議)。
Self-Discover: Automatically discover hardware/software changes in the network and populate the changes to the necessary subsystems to maintain always-sync state.
自我發現：自動發現網絡中的硬件/軟件更改，并將更改填充到必要的子系統中，以保持始終同步狀態。
Self-Config/Self-Organize: Whenever network changes happen, automatically configure corresponding hardware/software parameters such that the network is at the pre-defined target states.
自我配置/自我組織：每當網絡發生變化時，都將自動配置相應的硬件/軟件參數，以使網絡處于預定義的目標狀態。
Self-Monitor: Constantly monitor networks/services operation states and health conditions automatically.
自我監控：不斷自動監控網絡/服務的運行狀態和健康狀況。
Auto-Detect: Detect network faults, abnormalities, and intrusions automatically.
自動檢測：自動檢測網絡故障，異常和入侵。
Self-Diagnose: Automatically conduct an inference process to figure out the root causes of issues.
自我診斷：自動進行推理過程以找出問題的根本原因。
Self-Healing: Automatically take necessary actions to address issues and bring the networks/services back to the desired state.
自我修復：自動采取必要的措施來解決問題并將網絡/服務恢復到所需狀態。
Self-Report: Automatically communicate with its environment and exchange necessary information.
自我報告：自動與其環境進行通信并交換必要的信息。
Automated common operational scenarios: Automatically perform operations like network planning, customer and service onboarding, network change management.
自動化的常見操作場景：自動執行網絡規劃，客戶和服務入門，網絡變更管理等操作。
On top of those, these capabilities need to be across multiple services, multiple domains, and the entire lifecycle(TMF, 2019).
最重要的是，這些功能需要跨多個服務，多個域以及整個生命周期(TMF，2019)。
No doubt, this is the most ambitious goal that the networking industry has ever aimed at. It has been described as the “end-state” and“ultimate goal” of networking evolution. This is not just a vision on PPT, the networking industry already on the move toward the goal.
毫無疑問，這是網絡行業有史以來最雄心勃勃的目標。 它被描述為網絡發展的“最終狀態”和“最終目標”。 這不僅僅是關于PPT的愿景，網絡行業已經朝著目標邁進。
David Wang, Huawei’s Executive Director of the Board and President of Products & Solutions, said in his 2018 Ultra-Broadband Forum(UBBF) keynote speech. (David W. 2018):
華為執行董事兼產品與解決方案總裁王大衛在2018年超寬帶論壇(UBBF)主題演講中表示。 (David W.2018)：
“In a fully connected and intelligent era, autonomous driving is becoming a reality. Industries like automotive, aerospace, and manufacturing are modernizing and renewing themselves by introducing autonomous technologies. However, the telecom sector is facing a major structural problem: Networks are growing year by year, but OPEX is growing faster than revenue. What’s more, it takes 100 times more effort for telecom operators to maintain their networks than OTT players. Therefore, it’s imperative that telecom operators build autonomous driving networks.”
在完全連接和智能化的時代，自動駕駛已成為現實。 汽車，航空航天和制造業等行業正在通過引入自主技術來實現自我更新和更新。 但是，電信行業面臨著一個主要的結構性問題：網絡每年都在增長，但是運營支出的增長快于收入。 而且，與OTT運營商相比，電信運營商維護其網絡所花費的精力要多100倍。 因此，電信運營商必須建立自動駕駛網絡。”
Juniper CEO Rami Rahim said in his keynote at the company’s virtual AI event: (CRN, 2020)
瞻博網絡首席執行官拉米·拉希姆(Rami Rahim)在公司虛擬AI活動的主題演講中說： (CRN，2020)
“The goal now is a self-driving network. The call to action is to embrace the change. We can all benefit from putting more time into higher-layer activities, like keeping distributors out of the business. The future, I truly believe, is about getting the network out of the way. It is time for the infrastructure to take a back seat to the self-driving network.”
“現在的目標是建立自動駕駛網絡。 呼吁采取行動就是擁抱變化。 將更多的時間投入到更高層次的活動中，例如使分銷商脫離業務，我們都可以從中受益。 我真正相信，未來將使網絡暢通無阻。 現在該讓基礎架構在自動駕駛網絡中退后一步了。”
這個愿景可以實現嗎？ (Is This Vision Achievable?)
If you asked me this question 15 years ago, my answer would be “no chance” as I could not imagine an autonomous driving vehicle was possible then. But now, the vision is not far-fetch anymore not only because of ML/AI technology rapid advancement but other key building blocks are made significant progress, just name a few key building blocks:
如果您在15年前問我這個問題，我的回答將是“沒有機會”，因為我當時無法想象有可能駕駛自動駕駛汽車。 但是現在，這個愿景不再遙不可及，不僅因為ML / AI技術的飛速發展，而且其他關鍵構建塊也取得了重大進展，僅舉幾個關鍵構建塊：
- software-defined networking (SDN) control 軟件定義網絡(SDN)控制
- industry-standard models and open APIs 行業標準模型和開放API
- Real-time analytics/telemetry 實時分析/遙測
- big data processing 大數據處理
- cross-domain orchestration 跨域編排
- programmable infrastructure 可編程基礎架構
- cloud-native virtualized network functions (VNF) 云原生虛擬化網絡功能(VNF)
- DevOps agile development process DevOps敏捷開發流程
- everything-as-service design paradigm 一切即服務的設計范例
- intelligent process automation 智能過程自動化
- edge computing 邊緣計算
- cloud infrastructure 云基礎設施
- programing paradigm suitable for building an autonomous system . i.e., teleo-reactive programs, which is a set of reactive rules that continuously sense the environment and trigger actions whose continuous execution eventually leads the system to satisfy a goal. (Nils Nilsson, 1996) 適合建立自治系統的程序設計范例。 即遠程React程序，它是一組React性規則，可連續感知環境并觸發動作，這些動作的連續執行最終使系統達到目標。 (Nils Nilsson，1996年)
- open-source solutions 開源解決方案
巨大的挑戰 (Huge Challenges)
We have reasons to be optimistic about ADN while fully realize the considerable challenges in this ADN journey.
我們有理由對ADN感到樂觀，同時充分意識到ADN旅程中的巨大挑戰。
As we know, a typical autonomous system composes of 3 essential components:
眾所周知，一個典型的自治系統包含3個基本組成部分：
An Agent: A reactive system (controller) interacting with components of its environment so that specific goals are met.
代理：React性系統(控制器)與其環境組件進行交互，從而實現特定目標。
An Object: A physical or virtual component whose behavior can be controlled by system agents.
對象： A 可以由系統代理控制其行為的物理或虛擬組件。
The Environment: Consists of the elements of the physical and virtual infrastructure of the system that is used for the coordination between components (agents and objects).
環境：由系統的物理和虛擬基礎結構的元素組成，用于組件(代理和對象)之間的協調。
The first complexity of autonomous networking is a large number of objects (such as network elements, boards, links, and optical/electrical/wireless links). Each object has lots of tunable parameters and a lack of consistency of the object models. The other complexity is complex network environment: numerous networking protocols, operation models, layers of networking stacks, observability of the objects, communication infrastructures, and necessary interaction with the external environment like weather, sports events, and black swan events like current COVID19 pandemic. All these make building such an intelligent agent a very challenging task.
自主網絡的第一個復雜性是大量的對象(例如網絡元素，板，鏈路和光/電/無線鏈路)。 每個對象都有很多可調參數，并且對象模型缺乏一致性。 另一個復雜性是復雜的網絡環境：眾多的網絡協議，操作模型，網絡堆棧層，對象的可觀察性，通信基礎結構以及與外部環境(如天氣，體育賽事和黑天鵝事件，如當前的COVID19大流行)的必要交互。 所有這些使構建這樣的智能代理成為一項非常艱巨的任務。
Some other significant challenges come from the fact that telecom networks, like other infrastructure systems, evolve over a long time and not have many opportunities to be built “from scratch.” Instead, the new or changed elements are always needed to fit into the previously made infrastructure. Why is the realization of autonomous driving vehicles so much painful comparing with that of the autonomous driving airplanes? Because it has to fit into the existing complex and never perfect road infrastructure. Same for ADN, existing segregated, ultra-complex, and less-perfect network infrastructure presents significant challenges for the industry.
其他一些重大挑戰來自這樣一個事實，即電信網絡與其他基礎架構系統一樣，會長期演進，并且沒有很多“從頭開始”構建的機會。 取而代之的是，總是需要新的或更改的元素以適應先前制作的基礎結構。 為什么與無人駕駛飛機相比，實現無人駕駛汽車如此痛苦？ 因為它必須適合現有的復雜且永遠都不完美的道路基礎設施。 與ADN相同，現有的隔離，超復雜且性能較差的網絡基礎結構為行業帶來了巨大挑戰。
What is the strategy to get to ADN vision? As usual, the answer: divide, evolve, and conquer.
實現ADN愿景的策略是什么？ 和往常一樣，答案是：分裂，發展和征服。
自治網絡級別定義 (Autonomous Network Levels Definition)
As the autonomous driving vehicle, networking companies and industry groups also defined ADN evolution milestones.
作為自動駕駛汽車，網絡公司和行業組織也定義了ADN演進的里程碑。
TMF Autonomous networks levels (TMF, 2019)TMF自治網絡級別(TMF，2019)Level 0 — manual management: The system delivers assisted monitoring capabilities, which means executing dynamic tasks manually.
級別0-手動管理：系統提供輔助的監視功能，這意味著手動執行動態任務。
Level 1 — assisted management: The system executes a certain repetitive sub-task based on pre-configured to increase execution efficiency.
級別1-輔助管理：系統根據預先配置的內容執行某些重復的子任務，以提高執行效率。
Level 2 — partial autonomous network: The system enables closed-loop O&M for specific units based on the AI model under certain external environments.
級別2-部分自治網絡：系統在某些外部環境下基于AI模型為特定單元啟用閉環運維。
Level 3 — conditional autonomous network: Building on L2 capabilities, the system with awareness can sense real-time environmental changes. In certain network domains, optimize and adjust itself to the external environment to enable intent-based closed-loop management.
級別3-有條件的自治網絡：具有L2功能的系統可以感知實時的環境變化。 在某些網絡域中，優化自身并適應外部環境以啟用基于意圖的閉環管理。
Level 4 — high autonomous network: The system, building on L3 capabilities, enables, in a more complicated cross-domain environment, analyze and make the decision based on predictive or active closed-loop management of service and customer experience-driven networks.
級別4-高度自治的網絡：該系統基于L3功能，可以在更復雜的跨域環境中基于對服務和客戶體驗驅動的網絡的預測性或主動閉環管理來分析和制定決策。
Level 5 — full autonomous network: This level is the ultimate goal for telecom network evolution. The system possesses closed-loop automation capabilities across multiple services, multiple domains, and the entire lifecycle, achieving autonomous networks.
級別5-完全自治的網絡：此級別是電信網絡演進的最終目標。 該系統具有跨多個服務，多個域以及整個生命周期的閉環自動化功能，從而實現了自治網絡。
Cisco Digital Network Readiness Model (Cisco, 2020)思科數字網絡就緒模型(Cisco，2020) Juniper Self Driving Network Levels (K. Kompella, 2019)瞻博網絡自駕網絡水平(K.Kompella，2019)It is apparent that no standardized industry definition yet on the description of the levels. Different organizations have different focuses. But the main threads are similar:
顯然，關于級別的描述還沒有標準化的行業定義。 不同的組織有不同的重點。 但是主線程是相似的：
技術架構 (Technology Architecture)
A typical autonomous system has the following five complementary essential functions.
典型的自治系統具有以下五個互補的基本功能。
Perception, e.g., interpretation of stimuli, removing ambiguity from complex input data and determining relevant information;
感知，例如刺激的解釋，消除復雜輸入數據中的歧義并確定相關信息；
Reflection, e.g., building/updating a faithful environment run-time model from which strategies meeting the goals can be computed;
反思，例如，建立/更新一個忠實的環境運行時模型，從中可以計算出達到目標的策略；
Goal management, e.g., choosing among possible goals the most appropriate for a given configuration of the environment model;
目標管理，例如，在可能的目標中選擇最適合給定環境模型配置的目標；
Planning to achieve a particular goal;
計劃實現特定目標；
Self-awareness/adaptation, e.g., the ability to create.
自我意識/適應能力，例如創造能力。
The below diagram characterizes how these five elements interact with each other to achieve autonomy.
下圖描述了這五個元素如何相互影響以實現自治。
The Concept of Autonomy — Architecture Characterizations (Joseph S, 2019)自治的概念-建筑特征(Joseph S，2019)While this architecture very well applies to ADN conceptually, ADN has its unique complexities and need special considerations:
盡管此體系結構在概念上非常適用于ADN，但ADN具有其獨特的復雜性，并且需要特殊考慮：
Many Agents: Compare with a well-bounded, single-purpose system, i.e., robot/car/spaceship, networking more like a society. No single intelligent agent controls every aspect of the network but rather a collection of hierarchical smart agents that responsible for their subsystems, work in tandem.
許多特工 ：與功能強大的單一用途系統(例如，機器人/汽車/宇宙飛船)相比，網絡更像一個社會。 沒有單個智能代理控制網絡的每個方面，而是負責其子系統的一系列分層智能代理協同工作。
Intent-Driven: With the evolution to higher and higher levels in ADN, the human operators involve less and less in the actual network operation control loop. Instead, focus more and more on defining the business goal and processes “outside” or “on” the control loop. In order effectively to convey the business purpose to ADN, intent-driven interaction is one of the critical architectural considerations to separate the actual implementation from the request.
目的驅動：隨著ADN的不斷發展，人工操作員越來越少地參與實際的網絡操作控制回路。 取而代之的是，越來越多地專注于定義業務目標并在控制循環的“外部”或“外部”進行處理。 為了有效地將業務目的傳達給ADN，意圖驅動的交互是將實際實現與請求分開的關鍵體系結構考慮之一。
Centralized ML/AI Capabilities: In ADN, humans need a platform to generate/train/apply/optimize/share AI models. A centralized platform can make these work more efficient.
集中的ML / AI功能：在ADN中，人類需要一個平臺來生成/訓練/應用/優化/共享AI模型。 集中式平臺可以使這些工作更有效率。
Centralized Data Lake: Telecom network creates large amounts of data from various data sources. It is essential to have a centralized big data platform to collect, store, analyze, clean, filter data. Each subsystem can subscribe to the data it needs and make decisions by combining it with its local data.
集中式數據湖：電信網絡從各種數據源創建大量數據。 集中管理至關重要 大數據平臺收集，存儲，分析，清理，過濾數據。 每個子系統都可以訂閱其所需的數據，并通過將其與其本地數據結合起來進行決策。
Centralized Process Definition: Even though ML/AI will have the capability to decide the proper actions in particular scenarios, human operators still define the majority of the action sequence because not all the decision-making considerations are available for the machines to make a decision. A centralized process definition platform is highly desirable for operators to create/optimize/deploy workflows effectively.
集中的流程定義：即使ML / AI能夠決定特定場景下的適當動作，但是操作員仍然定義了大多數動作順序，因為并非所有決策因素都可用于機器做出決定。 對于操作員而言，非常需要集中式的流程定義平臺來有效地創建/優化/部署工作流。
ADN基礎架構 (ADN Infrastructure)
The networking industry and research communities have accepted the idea that the best approach to achieve ADN is through a set of single-domain autonomy plus cross-domain orchestration. Each single domain autonomy forms an AS(Autonomous System). Juniper calls the AS as “network bots” (K. Kompella,2018); others called it a “mini closed-loops.” This idea should have no surprise to everybody because it is very much in line with modern everything-as-a-service notations and famous cloud-native, microservice architecture. Powered by data, the AS possess more intelligence than regular microservices; it contains closed-loop control capabilities, understands the intent, performs the required operations, maintenance own health, and continuously monitor and adjust to ensure the target state is maintained.
網絡行業和研究界已經接受了這樣的想法，即實現ADN的最佳方法是通過一組單域自治和跨域編排。 每個單個域自治都形成一個AS(自治系統)。 Juniper將AS稱為“網絡機器人”(K. Kompella，2018年)； 其他人則稱其為“迷你閉環”。 這個想法對所有人都不應該感到驚訝，因為它非常符合現代的“一切即服務”概念和著名的云原生微服務架構。 由數據驅動，AS具有比常規微服務更多的智能； 它包含閉環控制功能，了解意圖，執行所需的操作，維護自己的健康狀況，并持續監視和調整以確保維持目標狀態。
The ASs run on an “ADN Infracture.” A good infrastructure lays a good foundation for ASs to thrive.
AS在“ ADN違規”上運行。 良好的基礎架構為AS的發展奠定了良好的基礎。
In the “ADN Solution White Paper, 2020”, Huawei outlines its ADN target architecture and product strategies.
在《 ADN解決方案白皮書，2020年 》中，華為概述了其ADN目標架構和產品策略。
Huawei ADN targeted architecture (Huawei ADN Solution White Paper, 2020)華為ADN目標架構(《華為ADN解決方案白皮書》，2020年)Huawei’s ADN target architecture covers all aspects of telecom networks: wireless, access, transport, optical, campus, and data centers. All three layers: NE, Network, and Cloud, embed with AI capability, which enables building ASs of different scales.
華為的ADN目標架構涵蓋了電信網絡的所有方面：無線，接入，傳輸，光纖，園區和數據中心。 三層：NE，網絡和云，均嵌入了AI功能，可構建不同規模的AS。
On product strategies, Huawei’s iMaster NAIE (Network AI Engine) acts as the AI platform for ADN. AUTIN (Automation Intelligent) business process definition platform to help carriers to build visualized, automated, and intelligent capabilities for operations. Meanwhile, MAE and NCE are mobile and fix network management, control, and analytics platforms.
在產品策略上，華為的iMaster NAIE(網絡AI引擎)充當ADN的AI平臺。 AUTIN(自動化智能)業務流程定義平臺可幫助運營商構建可視化，自動化和智能化的運營能力。 同時，MAE和NCE是移動的，可修復網絡管理，控制和分析平臺。
Huawei ADN System Panorama (Huawei ADN Solution White Paper, 2020)華為ADN系統全景圖(華為ADN解決方案白皮書，2020)This ADN infrastructure empowers AI anywhere with a clear demarcation of responsibilities. AI platform is especially important because of the skill gap among the architects who provide the design and requirements, the data scientists who built the models, and the software developers who develop and deploy end services.
這種ADN基礎結構可通過明確劃分職責的方式在任何地方為AI提供支持。 由于提供設計和需求的架構師，構建模型的數據科學家以及開發和部署最終服務的軟件開發人員之間的技能差距，AI平臺尤其重要。
In the Open Source community, Linux Foundation’s ONAP (Open Network Automation Platform) also targeted as the infrastructure to realize ADN. The gap of missing AI platforms has been addressed by several new initiatives from the Linux Foundation, one being Project Acumos, supported by AT&T.
在開源社區中，Linux Foundation的ONAP(開放網絡自動化平臺)也以實現ADN的基礎結構為目標。 Linux基金會的多項新計劃已經解決了AI平臺缺失的問題，其中一項是由AT＆T支持的Project Acumos 。
自治系統(AS) (Autonomous System (AS))
The many ASs work together in a defined ADN infrastructure to deliver overall ADN experience. How to identify and partition the responsibility of AS is a question that needs to consider. The AS can be packaged by technology, by region, by a legal entity, by functions, for re-use, for security, for simplified abstractions. Each company has its way of defining AS based on its own belief and available technologies.
許多AS在定義的ADN基礎結構中一起工作，以提供整體ADN體驗。 如何確定和劃分AS的職責是一個需要考慮的問題。 可以按技術，地區，法人實體，功能，重新使用，安全性，簡化抽象的形式對AS進行打包。 每個公司都有基于自己的信念和可用技術定義AS的方法。
If we want to achieve “open networking” for the benefits of sharing, simplifying, flexibility and cost-saving, standard organizations and open source communities can play significant roles in defining common ways to partition AS.
如果我們要實現“開放網絡”以共享，簡化，靈活性和節省成本的優勢，那么標準組織和開放源代碼社區可以在定義劃分AS的通用方法方面發揮重要作用。
The standard organizations are already making a move, e.g., ETSI published ZSM Requirements and Reference Architecture documents in 2019. ETSI defined a high-level architecture of ZSM. This ZSM framework would become the ADN infrastructure.
標準組織已經在采取行動，例如ETSI在2019年發布了ZSM需求和參考體系結構文檔。ETSI定義了ZSM的高級體系結構。 該ZSM框架將成為ADN基礎結構。
ZSM ArchitectureZSM架構ZSM’s perspective is to separate the system into Network Management Domains and E2E Service Management Domain. The domains connect via Domain Integration Fabric. Data Services are cross-domain. Base on the architecture, it further lists the set of services each domain should provide.
ZSM的觀點是將系統分為網絡管理域和E2E服務管理域。 域通過域集成結構連接。 數據服務是跨域的。 基于該體系結構，它進一步列出了每個域應提供的服務集。
ETSI GS ZSM 002)ETSI GS ZSM 002 )AS is a higher layer aggregation of those services with closed-loop capability. ETSI is in the process of defining resource closed-loop, network closed-loop, service closed-loop, business closed-loop, and user closed-loop. These overarching closed-loops fulfill the full lifecycle of the inter-layer interaction process, also identifying smaller closed-loop in each resource domain. Each closed-loop is an AS.
AS是具有閉環功能的那些服務的高層聚合。 ETSI正在定義資源閉環，網絡閉環，服務閉環，業務閉環和用戶閉環。 這些總體閉環完成了層間交互過程的整個生命周期，還識別了每個資源域中較小的閉環。 每個閉環都是一個AS。
Meanwhile, vendors also defined/productized various closed-loop components. For example, Juniper’s “Health Bot”
同時，供應商還定義/生產了各種閉環組件。 例如，瞻博網絡的“健康機器人”
Juniper HealthBot Closed-loop Automation Workflow (Juniper, 2020)瞻博網絡HealthBot閉環自動化工作流程(2020年6月)“看，我免提！” (“Look, I am hands-free!”)
A scary moment, isn’t it? How to earn the trust of the human operator to let go of the control to ADN? We can get some insights from an example.
可怕的時刻，不是嗎？ 如何贏得操作員對ADN放開控制權的信任？ 我們可以從一個示例中獲得一些見解。
Since 2018, Google has implemented a fully autonomous data center cooling system to improve operational efficiency and save cooling costs. The system works like this: every five minutes, the cloud-based AI pulls a snapshot of the data center cooling system from thousands of sensors. The AI system feeds the data into its deep neural networks algorithm, which predicts how different combinations of potential actions will affect future energy consumption. The AI system then identifies which actions will minimize energy consumption while satisfying a robust set of safety constraints with a re-enforcement algorithm. Those actions are sent back to the data center, where the suggested actions are verified by the local control system and then implemented. The system delivers an impressive consistent 30% energy saving, which translates to lots of dollars saving at Google scale over a long period. (Yevgeniy S., 2018)
自2018年以來，Google實施了完全自主的數據中心散熱系統，以提高運營效率并節省散熱成本。 該系統的工作方式如下：每五分鐘，基于云的AI從數千個傳感器中提取數據中心冷卻系統的快照。 AI系統將數據輸入其深層神經網絡算法，該算法預測潛在動作的不同組合將如何影響未來的能源消耗。 然后，AI系統識別出哪些動作將最大程度地減少能耗，同時通過強化算法滿足一組強大的安全約束。 這些操作將被發送回數據中心，在數據中心，建議的操作將由本地控制系統進行驗證，然后予以實施。 該系統可實現令人印象深刻的一致30％的節能，這意味著在Google范圍內可以長期節省大量資金。 (Yevgeniy S.，2018)
It is interesting to observe how human operators gradually give control to the autonomous system in the process.
有趣的是，觀察人員在此過程中如何逐步將控制權交給自治系統。
It started as an AI-based recommendation system; human operators took the suggestions and implemented them. This semi-automatic method delivered energy saving but also missed many power-saving opportunities. Because AI suggests a lot of fine-tuning operations to take advantage of the smaller changes in the environment, human operators can not afford to tune the system on all the granular actions at high frequency. The need for autonomous control arose. In 2018, Google completed the building of the closed-loop autonomous control system.
它最初是基于AI的推薦系統； 操作員采納了建議并予以實施。 這種半自動方法可以節省能源，但也錯過了許多節電機會。 由于AI建議進行許多微調操作以利用環境中較小的變化，因此人工操作員無法承受高頻下所有細粒度動作對系統進行的微調。 出現了對自主控制的需求。 Google在2018年完成了閉環自主控制系統的建設。
Google took a very cautious approach when introducing the autonomous system into the real world. First, they implemented a set of guardrails. Then they allowed the autonomous system to only control a small range of auto-tuning. Step by step, let the system gradually take over whole control. As described by Joe Kava, Google’s VP of data centers, “And you start to put in the guardrails to make sure that bad things can’t happen, and then you start to launch fully automated systems instead of semiautomated systems”(Yevgeniy S.,2018).
在將自治系統引入現實世界時，Google采取了非常謹慎的方法。 首先，他們實施了一系列的護欄。 然后，他們允許自治系統僅控制小范圍的自動調整。 逐步，讓系統逐步接管整個控制。 正如Google數據中心副總裁Joe Kava所描述的那樣，“然后您開始置入護欄，以確保不會發生壞事，然后開始啟動全自動系統，而不是半自動化系統”(Yevgeniy S. ，2018)。
The below chart shows the safety guardrails Google had implemented.
下圖顯示了Google已實施的安全護欄。
Google Autonomous Driving DataCenter Cooling System Safeguard Functions (Amanda G., et al., 2018)Google自動駕駛數據中心冷卻系統保障功能(Amanda G.，et al。，2018)The take away from Google’s experience is that every AS introduced in ADN evolution needs to put safety and robustness as the highest priority concern. We need to know the effect when things go wrong and have a mechanism to isolate the impact and able to put the network back to a known state. Google’s autonomous driving data center cooling control system sets a good example.
從Google的經驗中脫穎而出的是，在ADN演進中引入的每個AS都需要將安全性和魯棒性作為最高優先級。 我們需要知道出現問題時的影響，并且需要一種機制來隔離影響并能夠使網絡恢復到已知狀態。 Google的自動駕駛數據中心冷卻控制系統就是一個很好的例子。
Handing over control to AS is scary, and we should be scared. Critical infrastructure, like the network, has a massive impact on society if it goes down, not to mention monetary loss due to SLA violations. I believe it is a better strategy to hand over the control to AS gradually with fallback thought out on each step, like how Telsa does it in delivering the autonomous driving experience.
將控制權移交給AS令人恐懼，我們應該感到恐懼。 諸如網絡之類的關鍵基礎架構一旦崩潰，將對社會產生巨大影響，更不用說由于違反SLA而造成的金錢損失。 我認為將控制權逐步移交給AS是一種更好的策略，并在每個步驟上都考慮到后備問題，例如Telsa在提供自動駕駛體驗方面的做法。
The reason is straightforward, as being pointed out by Jeff Mogul of Google (Jeff M. 2018) regarding potential pitfalls of ADN:
正如Google的Jeff Mogul(Jeff M.2018)指出的那樣，原因很簡單，原因在于ADN的潛在陷阱：
- Any control system (human or automated) has its limits, 任何控制系統(人工或自動化)都有其局限性，
- Pushing a control system past its limits can cause crashes, 將控制系統推到極限之外可能會導致崩潰，
- Sometimes, the problem is not in the control system! 有時，問題不在控制系統中！
Jeff also argued that unlike autonomous driving vehicles where there is a complete stable manual-driven vehicle as its foundation. In networking, we do not have a perfectly stable manual operated network as our start point for ADN. We need to clean those unstable elements in the system before ADN can take off. “SelfDN success will depend on ‘fixing the environment,’ just as much as on great ML.” Jeff said.
杰夫還認為，與自動駕駛汽車不同的是，自動駕駛汽車具有完全穩定的手動駕駛汽車作為基礎。 在網絡中，我們沒有完美穩定的手動網絡作為ADN的起點。 我們需要先清理系統中的那些不穩定元件，然后才能起飛。 “ SelfDN的成功將取決于'修復環境'，以及出色的ML。” 杰夫說。
ADN作為社會技術系統 (ADN as a Socio-Techincal System)
Technology advancement dramatically increases the feasibility of meeting the technical challenges of an autonomous driving network. But if we just focus on meeting the technical requirements and ignore human, social, and organizational aspects, the ADN can be technically successful but operationally a failure and get rejected by society.
技術進步極大地提高了應對自動駕駛網絡技術挑戰的可行性。 但是，如果我們只專注于滿足技術要求，而忽略人員，社會和組織方面，則ADN在技術上可能是成功的，但在操作上可能是失敗的，并被社會所拒絕。
There is a need for a pragmatic approach to engineering ADN as socio-technical systems based on the gradual introduction of socio-technical considerations into existing hardware, software, human-machine interaction development processes.
在將社會技術考慮因素逐步引入現有硬件，軟件，人機交互開發過程的基礎上，需要一種實用的方法來將ADN作為社會技術系統進行工程設計。
In 2005, Playchess.com hosted a chess tournament in which teams of human players could use computer assistance during matches. The chess supercomputer Hydra was also entered into the competition. After recently defeating Grand Master Michael Adams 5 ?–? in a six-game match, it was considered the prohibitive favorite. Surprisingly, Hydra was eliminated before the semi-finals, with three of the four semi-finalists consisting of Grand Master-led teams equipped with supercomputers. Even more surprising was the fourth semi-finalist and eventual winner, team ZachS, composed of two relatively amateur chess players named Steven Crampton and Zackary Stephen, using ordinary computers.(Kyle B, John F, 2016)
在2005年，Playchess.com舉辦了一場國際象棋比賽，在該比賽中，人類運動員可以在比賽中使用計算機協助。 國際象棋超級計算機Hydra也參加了比賽。 在最近的六場比賽中以5 ?–?擊敗大師邁克爾·亞當斯(Michael Adams)之后，這被認為是令人望而卻步的。 令人驚訝的是，九頭蛇在半決賽之前被淘汰，四支半決賽中有三支由大師級領導的團隊組成，這些團隊配備了超級計算機。 更令人驚訝的是，第四名準決賽者并最終獲得冠軍的ZachS團隊由兩個相對業余的國際象棋手史蒂芬·克蘭普頓(Steven Crampton)和扎克里·斯蒂芬(Zackary Stephen)組成，他們使用普通計算機(凱爾·B，約翰·F，2016年)
The higher skill level of Hydra and the Grand Masters equipped with supercomputers was not enough to overcome the seamless collaboration between the less skilled amateurs and their weaker computers. As Garry Kasparov stated, “Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine +inferior process.” (Kyle B, John F, 2016)
九頭蛇和配備超級計算機的大師級別的較高技能水平還不足以克服技術水平較低的業余愛好者和較弱的計算機之間的無縫協作。 正如Garry Kasparov所說：“弱人機+更好的過程優于僅強大的計算機，更值得注意的是，優于強人機+劣等過程。” (Kyle B，John F，2016)
The take away from the chess story is that the human-machine combination has the potential to outperform human-alone and computer-alone. The same facts happen in many domains. For example, human forecasters at the National Weather Service can improve the accuracy of computer precipitation forecasts by 25% and computer temperature forecasts by 10% over computer-only predictions, and human-computer teams have the potential to outperform both doctors and computer algorithms at correctly interpreting mammograms. Human is good at spot outliner and can sense certain information that the machine does not have.
從國際象棋故事中脫穎而出的是，人機結合有潛力勝過單人和單機。 相同的事實在許多領域中也會發生。 例如，美國國家氣象局的人類預報員可以將計算機降水預報的準確性提高25％，將計算機溫度預報的準確性提高10％，而人機團隊則有可能超越醫生和計算機算法。正確解釋乳房X光照片。 人類擅長于輪廓描繪器，并且可以感知機器沒有的某些信息。
Socio-technical system design thinking is that the overall design objective is to optimize the system’s collective output efficiency, including organization, people, and technology.
社會技術系統設計思想是總體設計目標是優化系統的整體輸出效率，包括組織，人員和技術。
Socio-technical System社會技術系統While we are marching to ADN, socio-technical thinking, theory, and design methodologies are invaluable to ensure ADN a success.
當我們邁向ADN時，社交技術的思想，理論和設計方法對于確保ADN取得成功至關重要。
As an example, thinking about the future operating model of NOC/SOC (Network Operation Center/Service Operation Center), where is a prominent place that human operators and technologies interact in networking.
例如，考慮一下NOC / SOC(網絡運營中心/服務運營中心)的未來運營模式，其中人為運營商和技術在網絡中的交互作用最為突出。
In his article, “5 Overlooked Principles in the Race for Autonomous Networks”, contributed to TMF Insight, Yuval Stein from TEOCO wrote:
TEOCO的Yuval Stein在他的文章“自治網絡競賽中的5個被忽視的原則”中為TMF Insight做出了貢獻：
“ So long as controllers sit in the NOC/SOC, they will need to understand what’s happening in the network — even in areas where actions are occurring automatically. This means that, at least for the foreseeable future, the NOC will manage network problems with a mixture of manual and automatic resolutions. The needs of the NOC/SOC should continue to be considered — and respected.”
只要控制器位于NOC / SOC中，他們就需要了解網絡中正在發生的事情，即使是在自動執行操作的區域中也是如此。 這意味著，至少在可預見的將來，NOC將通過手動和自動解決方案的組合來管理網絡問題。 應該繼續考慮并尊重NOC / SOC的需求。 ”
He further elaborated:
他進一步闡述：
“NOC/SOC systems need to:
“ NOC / SOC系統需要：
Show relevant alarms at all layers. As noted, modern management systems are able to filter a large percentage of symptomatic alarms, but existing problems need to be listed.
在所有層顯示相關警報。 如前所述，現代管理系統能夠過濾大部分癥狀警報，但是需要列出現有問題。
Root-cause analysis is still required, even if the problem is resolved by a lower-layer management system. The NOC needs to have a clear, holistic view to understand what was resolved automatically — and what needs manual intervention.
即使通過較低層的管理系統解決了問題，仍然需要進行根本原因分析。 NOC需要有清晰的整體視圖，以了解自動解決的問題以及需要手動干預的問題。
Enrich alarms with organizational data that assists controllers. Information like geography, administrative ownership, and network relevant change requests must be associated to alarms.
利用有助于控制者的組織數據來豐富警報。 諸如地理位置，管理所有權和與網絡相關的變更請求之類的信息必須與警報相關聯。
Track the source of abnormalities, whether they are alarm-based or measurement-based. NOC controllers should understand the nature of the abnormalities as much as possible, which may require additional alarms, measurements or metadata.”
跟蹤異常源，無論是基于警報還是基于測量。 NOC控制器應盡可能了解異常的性質，這可能需要其他警報，測量或元數據。 ”
This kind of insight into the human operator’s needs when interacting with technology in future ADN is critical.
在將來的ADN中與技術交互時，這種對操作員需求的洞察至關重要。
摘要 (Summary)
Inspired by the rapid development of autonomous driving vehicles in recent years, ADN is the most ambitious goal in the networking industry ever. The formula of AI+Software Defined Network makes this goal seemly achievable. Vendors and standard organizations chart out the 6 phases of development, from mostly manual operation L0 to L5 of fully autonomous operation. The race to ADN already started, vendors not only publish white papers but also align their product strategies accordingly. Standard organizations also speed up the process of identifying requirements and defining the reference architectures.
受近年來自動駕駛汽車快速發展的啟發，ADN是網絡行業有史以來最雄心勃勃的目標。 AI +軟件定義網絡的公式使這一目標似乎可以實現。 供應商和標準組織列出了六個開發階段，從手動操作L0到完全自主操作的L5。 與ADN的爭奪已經開始，供應商不僅發布白皮書，而且相應地調整其產品策略。 標準組織還加快了確定需求和定義參考體系結構的過程。
Our current network infrastructure has stability issues that need to be addressed; in other words, cleaned up to set a good foundation for ADN.When designing ADN, we need to think beyond AI and networking technology; the socio-technical approach is essential.
我們當前的網絡基礎架構存在穩定性問題，需要解決。 在設計ADN時，我們需要思考的不僅僅是AI和網絡技術； 社會技術方法至關重要。
DARPA’s 2004 $1 million Grand Challenge for Autonomous Vehicle kicked off the rapid development of Autonomous Vehicle. Should we have such an event to kick start Autonomous Driving Network? K. Kompella, Juniper CTO, definitely thought so. He proposed the below “The Networking Grand Challenge”.
DARPA 2004年的100萬美元的無人駕駛汽車大挑戰賽開始了無人駕駛汽車的飛速發展。 我們是否應該有這樣的活動來啟動自動駕駛網絡？ 瞻博網絡首席技術官K. Kompella絕對是這樣認為的。 他提出了以下“網絡大挑戰”。
Self-Driving Network Grand Challenge (K. Kompella, Juniper 2019)無人駕駛網絡大挑戰(K.Kompella，瞻博網絡2019)Of course, we all know that telecom network like other infrastructure systems evolves over a long time and not have many opportunities to be built “from scratch”; instead the new or changed elements are always fitted into the previously made infrastructure. But this “clean-slate” exercise does have lots of value for validating our assumption, identifying the gaps and potential pitfalls in technology, and developing the new algorithms. Potentially this can spark some disruptive new technology. So I also think that is a good idea worth exploring.
當然，我們都知道，電信網絡像其他基礎設施系統一樣，是經過很長時間發展的，并且沒有很多“從頭開始”建立的機會。 取而代之的是，總是將新的或更改的元素安裝到先前制作的基礎結構中。 但是，這種“干凈的”練習對于驗證我們的假設，確定技術差距和潛在陷阱以及開發新算法確實具有很多價值。 這有可能引發一些破壞性的新技術。 因此，我也認為這是一個值得探討的好主意。
Mark W. (2014) Where to a history of autonomous vehicles.
馬克·W(Mark W.)(2014) 自動駕駛汽車的歷史 。
Nick F, Jennifer R (2017) Why (and How) Networks Should Run Themselves
Nick F，Jennifer R(2017 )網絡為什么(以及如何)自己運行
K. Kompella (2017) The Self-Driving Network: How to Realize It
K.Kompella(2017) 自駕車網絡：如何實現
K. Kompella (2018) The Self Driving Network: from vision to execution
K.Kompella(2018) 自我駕駛網絡：從愿景到執行
K. Kompella (2019) Self Driving Networks: Looks Mom, No Hands
K.Kompella(2019) 自我駕駛網絡：看起來媽媽，沒有手
Juniper (2020) Health Bot Overview
瞻博網絡(2020) 衛生機器人概述
Ciena (2019) Introducing the Adaptive Network Vision
Ciena(2019) 介紹自適應網絡愿景
TMF (2019) Autonomous Networks: Empowering Digital Transformation For the Telecoms Industry
TMF(2019) 自主網絡：助力電信行業的數字化轉型
Huawei (2018) Huawei’s David Wang: Moving Towards Autonomous Driving Networks
華為(2018) 華為的David Wang：走向自動駕駛網絡
Huawei (2020) Huawei ADN Solution White Paper
華為(2020) 華為ADN解決方案白皮書
Cisco (2019) Digital Network Readiness Model
思科(2019) 數字網絡就緒模型
David W. (2018) Moving toward Autonomous Driving Network
David W.(2018) 向自動駕駛網絡邁進
Huawei (2020) Huawei ADN Solution White Paper
華為(2020) 華為ADN解決方案白皮書
Cisco (2019) Cisco Digital Network Readiness Model
思科(2019) 思科數字網絡就緒模型
CRN (2019) Juniper Networks CEO: ‘The Goal Now Is A Self-Driving Network’
CRN(2019) 瞻博網絡首席執行官：``現在的目標是自駕車網絡''
Analysys Mason (2018) From Autonomous to Adaptive: The Next Evolution in Networking
Analysys Mason(2018) 從自主到自適應：網絡的下一個發展
Joseph S. (2019) Autonomous Systems — A Rigorous Architectural Characterization
Joseph S.(2019) 自治系統-嚴格的建筑特征
Timo P. (2018) The future of network and service automation
Timo P.(2018) 網絡和服務自動化的未來
Zeng W., et al.,(2017) Big data in telecommunication operators: data, platform and practices
Zeng W.，et al。，(2017) 電信運營商中的大數據：數據，平臺和實踐
Wylie W. (2017) Juniper Says its Bots are a Step Toward a “Self-Driving Network”
Wylie W.(2017) 瞻博網絡表示，其機器人是邁向“自動駕駛網絡”的一步
ETSI (2019) ZSM requirements ETSI GS ZSM 001
ETSI(2019)ZSM要求ETSI GS ZSM 001
ETSI (2019) ZSM reference architecture ETSI GS ZSM 002
ETSI(2019)ZSM參考架構ETSI GS ZSM 002
Yevgeniy S. (2018) Google is Switching to a Self-Driving Data Center Management System
Yevgeniy S.(2018) Google正在切換到自動駕駛數據中心管理系統
Amanda G., et al., (2018) Safety-first AI for autonomous data center cooling and industrial control.
Amanda G.，et al。，(2018) 安全第一的AI，用于自主數據中心散熱和工業控制。
Jeff M. Unsafe at Any Speed: Self-Driving Network without Shelf-Crushing
杰夫·M(Jeff M.) ：任何速度都不安全：無人駕駛的自動駕駛網絡
Yuval S. 5 overlooked principles in the race for autonomous networks
Yuval S.5 在自治網絡競賽中被忽略的原則
Nils Nillson, 1996 Introduction of Teleo-Reactive Programs
Nils Nillson，1996年，采用Teleo-Reactive程序
Kyle B, John F, 2016 Autonomous Systems to Sociotechnical Systems: Designing Effective Collaborations
Kyle B，John F，2016年從自治系統到社會技術系統：設計有效的協作
翻譯自: https://medium.com/swlh/autonomous-driving-network-adn-on-its-way-772d053f8a1e
自動駕駛發展
總結
以上是生活随笔為你收集整理的自动驾驶发展_自动驾驶网络及其发展的全部內容，希望文章能夠幫你解決所遇到的問題。
- 上一篇： 诺贝尔奖距离中国内地多远(ZT)
- 下一篇： Axure交互之链接行为