Research On Collaborative Monitoring Of Hierarchical Heavy Hitters Based On SDN

Network measurement provides necessary information for network management applications and plays an essential role in ensuring the stability and security of the network.Hierarchical Heavy Hitters,one of the network measurement tasks,is presented to identify Heavy Hitters that aggregates based on common IP prefixes and has network applications such as anomaly detection and DDo S detection.Many existing studies use the centralized algorithms on a single switch to identify Hierarchical Heavy Hitters among all the network flows through it.However,the continuous expansion of the network traffic scale makes it difficult for those algorithms to achieve the expected measurement accuracy.In the measurement of Hierarchical Heavy Hitters,the error of lower levels will affect the accuracy of high levels,which is called the dependency problem in this thesis.Due to the limited resources and dependency problem,the accuracy of Hierarchical Heavy Hitters measurement on a single switch is very low.Benefit from Software-Defined Network(SDN),Software-Defined Measurement(SDM)is proposed,which enables a distributed way to deploy the measurement tasks on multiple switches for the collaborative monitoring of flow efficiently.However,in such a distributed environment,the small sub-flows(with a common IP prefix)from a large flow are distributed on different switches for measurement,which will be ignored as they are too small,resulting in aggregation at a higher level,which is called the problem of missed sub-flows in this thesis.In order to realize SDN-based collaborative monitoring of Hierarchical Heavy Hitters in a distributed environment,this thesis first formulates the problem of strategy formulation as a Joint Optimization of Task Deployment and Flow Selection problem(JOTDFS),and solves it approximately.The purpose is to use the resources of the network-wide switches for measurement and alleviate the problem of low accuracy caused by dependency problem and limited resources.Then,this thesis proposes a Distributed Framework for Hierarchical Heavy Hitters Measurement(DF3HM)to alleviate the problem of missed sub-flows.Under this framework,each measurement task deployed on a switch has a compact data structure,which consists of a Bloom filter,a Heavy part,and a Light part.Bloom filter is used for the collaborative monitoring of Hierarchical Heavy Hitters on multiple switches according to the strategy of JOTDFS problem,Heavy part is used to maintain the identifiers and count values of potential large flows,and Light part only records the count values of small flows for data recovery.Experiments based on the CAIDA dataset verify the effectiveness of the approximate solutions to the JOTDFS problem and the framework DF3 HM,which implies that they can balance loads of switches more effectively and improve the accuracy of Hierarchical Heavy Hitters measurement.This thesis also prototyped DF3 HM with different kinds of combined data structures by P4 language and verified the feasibility of deployment on the switch.