Research On Key Technologies Of Intent-Driven Network Automated Management

With the increasing complexity of network service requirements and the continuous expansion of network scale,the difficulty of network management is increasing.The traditional network management model that relies on manual configuration is hard to satisfy with the high requirement of agility for network controlling and management in the current business.In recent years,academia and industry have proposed a series of new technologies,such as software-defined networking,network function virtualization,and microservices,to bring flexibility and agility to the solid network management.Although these technologies can simplify network device management,most of the network parameters still need to be configured in manual,which cannot ensure an efficient and reliable network management.Intent-driven networking(IDN)is proposed as a new paradigm for network management.It places business requirements(i.e.intents)at the core of network service delivery,and uses a decoupled layered network control model and closed-loop network orchestration to achieve intent-driven network automation.IDN mainly includes three key technologies: intent description,intent translation(from intents to configuration),and business assurance(verification of the consistency between network and intents,and intent optimization/repair).In accordance with the input intents,it can automatically generate,verify,optimize and deploy configurations to achieve the desired network state,and thus can replace the traditional manual network management.IDN is the innovation of network management and the synthesis of a series of network automation technologies.However,IDN does not yet have a unified standard,and relevant network architectures and technical supports need further study.Hence,there are still many challenges to realize an intent-driven network automated management.An intent can be divided into functional and non-functional: the former has clear operational requirements for network management,and can be implemented through the mapping relationship between intent semantics and configuration template;the latter refers to network attributes and involves complex and diverse management operations,like adjusting and optimizing network policies according to various factors such as the uncertainty of networks,or comprehensively perceiving the state space of network behaviors to verify whether the network meets this attribute,which pose many challenges to an automated implementation of intents.To this end,in this thesis,we focus on the implementation of nonfunctional intention and conduct our research from three aspects:the translation method for coexisting but conflicting intents,consistency verification between protocol logic behavior and state reachability intents,and reliability guarantee of intention implementation.Our main research contents and achievements are as follows:1.The first issue of intent implementation is how to translate abstract intent into network configurations.A single intent can be converted using the mapping relationship between the intent and network resource.However,during the practical network service provisioning,multiple business intents can coexist and even conflict with each other,and thus,the translation via relationship mapping will fail.When these intents are non-functional and compete with each other in terms of resources,the intent translation will face great challenges.Most of the existing methods only consider the translation of a single non-functional intent.In this thesis,we studied the problem of multi-intent-driven service orchestration under the microservice architecture.We point out that low energy consumption,high quality of service and system load balancing are the three key but conflicting non-functional intents due to the splitting of network elements,and the translation method for these three intents needs a more finegrained resource model to ensure the effectiveness of the generated orchestration policies.To this end,we construct a fine-grained system resource model for microservice orchestration,and propose a multi-stage intent translation algorithm based on multi-objective optimization.Extensive simulation results show that the service orchestration decisions generated by our designed algorithm can satisfy with multiple conflicting business intents to the greatest extent without introducing additional calculation time overhead.2.Verifying whether the network behaviors satisfy the intent is another key to realize the intent implementation.The most important factor to define the network behaviors is the logic of the used network protocols itself.The non-functional intents of state reachability are needed to verify the correctness and security of the protocol behavior logic,so as to ensure the reliable network operation.However,network protocols are constantly evolving and updating,and the resulting complex state space makes it difficult for existing detection tools to achieve an automated consistency verification process.To address the verification problem of the consistency between protocols and state reachability intents,in this thesis,we take cellular network protocols as an example and propose an automatic consistency verification method based on formal methods.Firstly,we focused on automatically constructing protocol models from network software and optimizing the state space of constructed models.Then,we proposed a model checking algorithm based on counterexample-guided abstraction refinement and sparse value flow strong update.Finally,we designed a series of state reachability intents,and conducted practical test and simulation network validation.We found many unknown logic flaws in cellular network protocols with our proposed method.These flaws cannot satisfy the desired intent and can involve abnormal network operations and even security risks.3.IDN adopts a decoupled and layered network control model,and its closed-loop intent implementation process relies on the support of the control plane which acts as an intermediate node.Therefore,the correctness of the control plane directly affects the reliability of intent implementation.The control plane uses module-customizable software to implement the control logic required for intent implementation.Considering various types of asynchronous and concurrent input events,these logics are usually designed to be non-deterministic,which brings complex analysis processes to failure diagnosis.Most of the existing research methods use black-box testing to figure out a set of input events that can trigger the abnormality of the control plane.However,the root cause of the failure in the control plane still needs to be further investigated,which limits the rapid repair of failures.To this end,in this thesis,we analyze the characteristic of failures occurred in the software implemented logic in the control plane,and propose a lightweight control plane behavior monitoring mechanism,a context-aware system behavior model construction algorithm,and a failure location algorithm based on differential checking.With these proposed technologies,our failure diagnosis system is realized.Experiments show that our proposed system can effectively diagnose failures in the control plane within a low performance effect,and thus,can be applied to practical network operations.