Research On Adaptive Network Monitoring Technology Based On In-band Network Telemetry

Recently,the rapid development of network technology has made the network services and infrastructure equipments become increasingly complicated,which leads to the dilemma of the traditional network control and management(NC&M).Thus,the researchers turn to propose technologies such as software defined networking(SDN)and in-band network telemetry(INT)to provide real-time and fine-grained network monitoring.On the one hand,SDN decouples the control plane and the data plane,and provides logically centralized control over the data plane,which enables the NC&M to achieve full of flexibility and programmability.On the other hand,INT guarantees packet-level end-to-end network monitoring by inserting telemetry data into a flow’s packets,so that NC&M can deeply visualize the real-time end-to-end experience of every packet in the network.However,in the dynamically high-speed network,INT scheme still faces many challenges during the practical deployment,such as flexible,efficient and dynamic deployment issues.Firstly,the programming protocol-independent packet processors(P4)based per-packet INT monitoring involves in high overhead and it is hard to efficiently support the runtime adjustment of the pipeline,resulting in the performance bottlenecks and flexible deployment issues.Secondly,although sampling based INT scheme can effectively reduce the monitoring overhead,there is still no related works to further consider how to efficiently balance information gain and monitoring overhead for sampling based INT orchestration in the multi-telemetry task scenario.Finally,the static INT scheme cannot adapt to the dynamic network status timely,thus it will be no longer reasonable due to the fluctuation of network state,which probabilistically leads to dynamic deployment issues.Therefore,in response to the above challenges,this dissertation proposes research on adaptive network monitoring technology based on in-band network telemetry,which covers the solutions of the static and dynamic deployment problems.We summarize the specific research topics and contributions below:·As for the flexible deployment issues of INT scheme in the dynamic network,especially the shortcomings of runtime programmability and monitoring overhead of P4-based INT scheme,we propose and design a runtime programmable and low-overhead selective INT scheme(Sel-INT)based on protocol oblivious forwarding(POF).In this dissertation,we first time extend Open vSwitch(OVS)to make it POF-enabled,and obtain a brand-new software-based POF switch(OVS-POF),which can process packets in a protocol-agnostic way without severe performance degradation.Then,we leverage POF and select group table to implement Sel-INT on OVS-POF,which is a protocol-agnostic and runtime programmable network monitoring system.Moreover,we present the specific mechanism to runtime adjust the Sel-INT pipelines on the data plane,and the reasonable approach to balance the monitoring accuracy and sampling ratio on the control plane.We finally realize the flexibly runtime programmable network monitoring system,which not only demonstrates that Sel-INT can dynamically program the pipelines of INT scheme,but also confirms that our proposal can ensure proper accuracy and timeliness for network monitoring while greatly reducing the overheads of INT.Therefore,we improve the flexible deployment capability of INT scheme in the dynamic network.·As for efficient deployment issues of INT scheme in the dynamic network,especially the lack of reasonable deployment strategies for sampling based INT scheme,we propose corresponding network-level monitoring orchestration algorithm for multi-telemetry tasks in Sel-INT scenarios.In this dissertation,we first time study how to use Sel-INT wisely in a network such that the tradeoff between monitoring accuracy/coverage and INT overheads can be balanced well.Specifically,we propose a Lagrangian based polynomial-time approximation algorithm to solve the orchestration problem about the INT metadata selection approach and the INT metadata assignment strategy on the data plane.We finally realize the network-wise orchestration and efficient monitoring of Sel-INT scheme,which not only provides a near-optimal feasible solution,but also effectively balances the information gain of INT and monitoring overhead of INT.Therefore,we improve the efficient deployment capability of the INT scheme in the dynamic network.·As for dynamic deployment issues of INT scheme in the dynamic network,especially the neglect of dynamic strategies in the researches such as the optimization and application of INT scheme with the fluctuant network state,we propose the actual operation scheme for self-adaptive network monitoring based on Sel-INT intelligently.In this dissertation,we first combine the advantages of POF and P4 to build a generic programmable data plane(G-PDP)as well as dual-pipeline programming for Sel-INT on G-PDP.We also propose the difference based filtering mechanism and machine learning assisted analytics method,in order to accelerate the real-time processing of massive telemetry data.For the flow-level monitoring scheme and link-level monitoring scheme in the IP layer and EON layer,we leverage Sel-INT to design and implement self-adaptive network monitoring decision-making schem es on the control plane,respectively.Moreover,we leverage information entropy to realize a local decision-making scheme to focus on the collection of valuable telemetry data on the data plane.We finally realize the self-adaptive network monitoring operation schemes according to the network state and real-time data analytics,which ensures the real-time,correct and feasible network monitoring.Therefore,we improve the dynamic deployment capability of the INT scheme in the dynamic network.