| The "static" and "rigid" traditional network architecture can hardly respond to users’ service demands quickly.In response to the above problems,the Smart Integration Identifier Networking based on "three-layer" and "three-domain" architecture decouples service space and network space,constructing a Service Function Chain(SFC)composed of Virtual Network Function(VNF)by innovatively introducing a knowledge domain to guide the unified scheduling and integration of network-wide resources to improve service deployment efficiency and provide high-quality network services to users.However,the dynamic change of SFC orchestration in the Smart Integration Identifier Networking increases the difficulty of network management.For one thing,the characteristic of VNFs sharing underlying physical resources easily causes network service resource contention,which increases the risk of hardware and software failure and degrades network performance;for another,the complex connection between VNFs and underlying hardware devices makes fault propagation model difficult to establish accurately and enhances the difficulty of fault detection.To address the above problems,this dissertation proposes a fault detection mechanism for service chains based on the Smart Integration Identifier Networking architecture,aiming to achieve fine-grained sensing of the full range of SFC states with minimal overhead and improve the efficiency and accuracy of fault detection.The main work of this dissertation is as follows:(1)Proposing a hybrid sensing mechanism for the SFC state.The Rest API technology is used in the resource adaptation layer,and the in-band network telemetry technology is used in the network component layer to collaboratively sense the service function state and network state,to realize fine-grained active sensing of SFC heterogeneous resources with a global perspective and expand the scope of SFC state sensing.(2)Proposing a detection path optimization mechanism for SFC fault detection scenarios.Considering the telemetry overhead and the real-time requirements of network status data,a depth-first search algorithm is used to plan a detection path that simultaneously satisfies the three conditions of length balance,covering the SFC network and no overlapping links,to reduce the telemetry overhead and telemetry delay difference at the network component layer.(3)Proposing a convolutional long and short-term memory network-based SFC fault detection algorithm.Due to the high-dimensional and complex characteristics of largescale SFC state data sensed by multiple sources,the convolutional layer and the longshort memory network layer are used to extract local fault features and mine the correlation between long-distance fault features,respectively,to improve the accuracy of SFC fault detection.To verify the feasibility and effectiveness of the proposed mechanism,the dissertation deploys six SFCs in a prototype system based on Smart Integration Identifier Networking and simulates the occurrence of SFC faults by fault injection for functional and performance testing.The experimental results show that the proposed service chain fault detection mechanism can realize the sensing of the full range of SFC states with a small overhead,and at the same time,the efficiency and accuracy of SFC fault detection are improved.Among them,the detection overhead is reduced by about 65.7% compared with the path-tracer detection method;the network sensing delay difference is reduced by about 94.4% compared with the SFC passive sensing scheme;the accuracy of the proposed SFC fault detection algorithm is 96.98%,and it can effectively identify ten SFC state types from the unbalanced data compared with the convolutional neural network. |
