| The core of the network management mechanism is to monitor network devices and then send operating instructions based on the monitoring results.Traditional distributed network management uses SNMP,Netflow and other network management methods,which have gradually failed to meet the requirements in response to the growing network scale.For example,in the data center network,the traditional network management method cannot adapt to the large-scale,low-latency,and unpredictable characteristics of the data center network.Under the software-defined network architecture,the emergence of the separation of relay and control and the programmable data plane makes the in-band network telemetry(INT)capable of low-latency monitoring play an advantage.Therefore,how to use in-band network telemetry under the SDN architecture to build a highly available and highly reliable network management architecture to meet fine-grained monitoring,fault detection and location,and error feedback is a worthy research topic.However,in the current SDN network management research,there is a lack of open source solutions for highly available network management,and it is difficult to meet the requirements on the granularity of network management equipment monitoring.With the development of SDN,the concept of data plane monitoring,INT,only regulates simple data transmission formats and primitives at the underlying device level,and has not standardized the implementation of network monitoring architecture.Therefore,this paper builds a complete network management architecture(HAMon)based on INT,which realizes fine-grained network monitoring that guarantees high availability,and has considerable advantages in monitoring performance.It mainly includes the design and implementation of the following four contents:(1)Independent high-availability management planeHAMon extends the traditional SDN architecture and creates an independent management plane.The management plane is mainly responsible for collecting INT messages transmitted by the data plane and storing the results of message analysis.A high-speed packet processor Collector implemented by XDP is specifically designed,and a time-series database cluster supporting high-performance reading and writing is constructed.(2)Large-scale data plane supporting INT functionThe HAMon data plane optimizes the P4 table item issuing mechanism,designs the table item issuing method,and uses the publish/subscribe mechanism to monitor the upper-level control instructions in a timely manner.At the same time,the INT transmission message format is designed and implemented in the data plane using P4.(3)High-availability algorithm MPS for management plane clusterHAMon is designed as a result of multiple management planes,using a clustered deployment method.The MPS algorithm is aimed at the load balancing of the management cluster.The master-slave election and backup disaster recovery respectively correspond to the MPS-LB strategy,MPS-Election algorithm and MPS-DT scheme respectively,ensuring the high availability of the management plane cluster.(4)Monitoring path balanced division algorithm BPOGThe BPOG algorithm is implemented in the HAMon control plane to provide guidance for the behavior of INT probes in the data plane.The BPOG algorithm uses a heuristic simulated annealing algorithm to focus on the balanced division of network links.It is better than the existing network monitoring path division method in terms of the flexibility and effect of path division.In general,the HAMon architecture designed in this paper constructs a complete set of highly available network management architecture through the highly available management plane combined with the management module in the control plane to coordinate the data plane.HAMon has a great advantage in monitoring the link coverage of the entire network,achieving fine-grained monitoring with low load and high accuracy.At the same time,HAMon can locate and report the fault quickly at a rate of microseconds when a link failure occurs. |
PDF Full Text Request