Unified Orchestration and Distributed Management and Control for Network Automation and Intelligentization

 

 

Developments in cloud computing, AI and 5G are driving cloud-network convergence. Network operation and maintenance (O&M) in the cloud-network convergence era will face the challenges of complicated network forms including multiple network domains and multiple cloud platforms, diversified industrial application services, and management of ultra-large-scale network connections. As a result, the new O&M mode must be automated and intelligent.       

A Hierarchical Architecture 

ZTE proposes a hierarchical management and control architecture (Fig. 1) after analyzing the complicated cloud-network convergence O&M scenarios while taking into consideration of the overall development trends of the industry.   

The business logic layer at the top is the business application layer of cloud-network convergence. Apps at this layer represent specific business application scenarios and can capture and classify business intents of end users. 

The service orchestrator layer needs a unified cloud-network convergence orchestrator. The service orchestrator can translate a business intent into specific service logic of cloud-network convergence, and further translate the service logic to network orchestration requests and NFV service chain orchestration requests, which will be finally sent to the network orchestrator. With unified E2E orchestration of networks and clouds, this layer centrally represents cloud-network convergence.  

The network orchestrator layer is interconnected with the service orchestrator layer. SDN orchestration refers to orchestrating SDNs across multiple domains to offer E2E connections. The NFV orchestrator (NFVO) performs unified orchestration of VNFs. The unified cloud management system uniformly manages and schedules various cloud platforms.

The distributed management and control layer is the layer for manufacturers' controllers. It allows the MAN and the backbone network to use different SDN controllers. Multiple SDN controllers can be used in the MAN to realize network decoupling. In the cloud computing scenario, a VIM cloud management platform can be deployed for a telecom cloud and an independent cloud management/scheduling system for a public cloud at the control layer.  

Unified Orchestration, Distributed Management and Control

In view of the above-mentioned architecture, ZTE believes that unified orchestration and distributed management and control are the key architecture for the O&M in the new cloud-network convergence era.

Distributed Management and Control for Network Efficiency and Reliability

Distributed management and control means deploying a controller or management system in each management and control domain of the cloud network to control and manage the domain's networks and cloud platforms, which not only guarantees the agility and efficiency of each domain but also ensures high reliability of the whole system.

A WAN spans a large geographic area. If a management and control system is deployed centrally, the system will need to access the widely distributed network devices, which reduces the network reliability between the network devices and the management and control system and hinders efficient network control and scheduling. The current network forwarding plane also has some distributed control functions, which need to cooperate with the centralized control functions of the management and control system.   

Each management and control domain is generally built by one equipment manufacturer. With the continuous development of network technologies, southbound interfaces are becoming more diversified and complicated. There are at least hundreds of southbound interface-based service models. Only the management and control system provided by the corresponding network equipment manufacturer can provide more efficient and reliable southbound interfaces. 

Unified Orchestration for E2E Cloud Network Service Delivery  

Unified orchestration refers to connecting the management and control system of each domain at the orchestrator layer and providing unified orchestration interfaces of cloud-network convergence. It has the following advantages:

—Differences among network domains and cloud platforms are hidden. The distributed management and control systems are uniformly connected to the centralized orchestration system, and the system interfaces from different vendors are standardized. This gives full play to the innovative features of different manufacturers' products. 

—Service-oriented interfaces are easy to standardize. Different from the management and control system that is oriented to network technology, the interfaces at the orchestrator layer are oriented to services and business intent, and can be abstracted and standardized according to service models. 
—The evolution of lower-layer network technologies has much less influence on the upper-layer service system. Through abstraction implemented via the orchestration layer, many technical details of the lower-layer network are hidden, which not only allows lower-layer network vendors to make technology innovations but also improves service system stability.   

Therefore, unified orchestration is an inevitable choice of cloud network service E2E delivery. 

Hierarchical Automation and Intelligentization of Cloud Network Services 

Based on the architecture for unified orchestration and distributed management and control, cloud network services can be automated and intelligentized hierarchically. The distributed management and control layer is closely connected with network devices, and based on innovative technologies of different vendors, it enables intelligentization and local closed-loop automation. The unified orchestration layer performs business logic oriented E2E service orchestration to implement "customer intelligence" and E2E service automation.

Hierarchical automation and intelligentization of cloud network services can be analyzed from the following aspects. 

Intelligent Proactive Data Collection

Real-time network perception is the basis of network automation, and real-time collection technology is key to real-time network perception. Traditional minute-level data collection via SNMP or FTP cannot meet the requirement of real-time collection. It is recommended to use telemetry collection. Telemetry supports millisecond-level hardware sampling, thus providing very high collection efficiency. However, it brings 100 times more data than the traditional collection modes while increasing collection frequencies, which makes data storage and analysis much more difficult. 

ZTE developed its patented technology of intelligent telemetry collection based on AI technology. This technology builds an optimal learning model through active AI learning to realize low-density collection of steady-state data and high-density collection of bursty data. It also makes scientific predictions based on AI, which greatly reduces the data volume to be collected and stored, and substantially increases the real-time data analysis efficiency.

Hierarchical Big Data Analysis 

Big data-based service analysis is the foundation for an intelligent network. The value of hierarchical big data analysis is that it better aligns with the analytical requirements at each level and provides analytical capabilities that are more appropriate to the application requirements.

The distributed management and control system integrating subnet analysis modules in the domain provides subnet-level analysis, and with a focus on real-time analysis in the domain, it drives real-time network protection based on AI policies. It provides real-time analysis policies for intelligent fault diagnosis and performance degradation of networks inside the domain. The analysis capability required at this layer is mainly the real-timeness of analysis responses. 

The unified orchestration layer integrating an independent big data platform can identify customers precisely and provide proactive analysis and protection in view of business logic and service quality, which improves customer satisfaction. This layer may have lower requirements for real-time responsiveness but should support mass data storage and analysis, offline history big data analysis, parallel technology and analysis, and AI-based analysis. 

Hierachical Closed-Loop Automation

Unified policy management and hierarchical automatic closed loops are the basis of dynamic cloud network service automation.

Policies are the basis of automatic closed-loops. Unified policy management requires centralized policy management, which ensures unified scheduling of network automation polices and also reduces policy repetition and complexity.

Automatic closed loops at the distributed management and control layer support self-healing of services inside the domain and second/millisecond-level service recovery. At the same time, distributed automation policies can prevent adverse chain effects brought by local service interruption that may cause whole network failure. 

Automatic closed loops at the unified orchestration layer can employ preventive service protection and provide minute-level service switchover and self-healing.

Automated E2E SR Path Creation

An E2E segment routing (SR) path of the cloud network service generally needs joining multiple manufacturers' SR paths. Hierarchical SR path creation can use the following way:

The unified orchestrator computes the start and end nodes of the E2E SR path and the overall SLA quality requirement for the SR path according to the service requirements. E2E quality detection is performed for the SR path. If the overall quality requirement meets the service requirement, SR segmentation requests (including the path information and the split SLA indicators) will be sent to each distributed management and control system. 

The distributed management and control system computes the optimal local SR path inside the domain according to the requirement of the orchestrator and reports the path information to the unified orchestrator. 

Hierarchical SR path creation makes full use of the advantages of the unified orchestrator in centralized service logic processing and E2E management as well as the optimal path computation capability of the distributed management and control system.

Cloud-network convergence is the development trend and the core competitiveness of operators. Facing the requirement of unified E2E management and control and orchestration of various domains and cloud platforms, ZTE believes that unified orchestration and distributed management and control are the optimal O&M architecture to realize cloud network service automation and intelligentization.