Research On Intelligent Flowtable-Update Mechanism In Software-Defined Datacenter

Software-Defined Datacenter(SDDC)utilizes the idea of “software-defined everything” and virtualizes network computing resources and storage resources,therefore,to support fine-grained and efficient management and scheduling of network services.Software-Defined Networking(SDN)is one of vital components of SDDC,and the flowtable in each SDN switch plays a crucial part in the implementation of SDN routing and forwarding strategies,which supports and guarantees the traffic management and network services in SDDC.While the network traffic is at its peak,that is to say explosive and instantaneous short-lived flows are generated and transmitted,and the storage resources of SDN switches are so limited that the forwarding policies(i.e.flow entries)will rapidly consume the flowtable storage of SDN switches: Ternary Content Addressable Memory(TCAM),even worse,these resources cannot be reclaimed in time.Finally,the storage space of flowtables will encounter a bottleneck,and the efficiency of packet forwarding in SDDC will decrease sharply.Aiming at the above issues of flowtable storage and packet forwarding,this thesis proposes an Intelligent flowtable-update mechanism,named Multi-staged Eviction Mechanism or MEM,which mainly optimizes the management of flowtable from two aspects:(I)flow entry classification,and(II)flow entry prediction and update.Due to the differences in the flow entry management by the diversified flows,a method for classifying flow entries based on flow characteristics is proposed in this thesis.Because large amounts of flows in SDDC present different patterns,this method consists of two stages,and flow entries are classified based on the flow persistent state and some statistical information.The first stage focuses on and analyzes the flow persistent state in SDDC,and classifies periodic flow entries into disposable and retainable flow entries.However,this rough classification in the first stage cannot optimize and granularly manage the corresponding flows of flow entries.In the second stage,the flow entries from the first stage will be normalized,and a flow entry decision tree(FETA)is established after maximum optimal features are selected according to the flow statistics intelligently.In order to manage flow entries of same mode,the two-staged classification ultimately classified these flow entries into disposable and retainable flow entries.A flowtable update mechanism based on ARMA(Autoregressive Moving Average)prediction(FUMA)model is proposed,taking the temporality of flow entry survival states into consideration and performing after the two-staged classification.For the disposable flow entries generated by the periodic short-lived flows in the classification stage,the proposed FUMA predicts the survival state of these flow entries using their historical survival states in the next sampling period.And then,some predicted flow entries with less frequently used and shorter survival period will be evicted according to the flowtable storage utilization.In the next sampling period,the predicted result is adaptively applied as the MEM feedback for the management optimization of flow entries,so as to implement intelligent update of flowtables.MEM implements the intelligent update of flowtables by its self-judgement and adaptive management on the corresponding flow states,which finally reduces the stress of flowtable storage and increases the packet forwarding rate.Finally,Simulation experiments have been conducted on two datacenter datasets to evaluate the impact of applying MEM on the flow entries and network performance.The results show that the proposed MEM outperforms the mechanism of predicting the number of flow entries based on the original ARMA,in terms of the flow entry compression ratio reflecting the storage utilization and the packet forwarding ratio which affects the quality of traffic interactions.The proposed method of two-staged classification works well in flow entry classification while there are a large number of short-lived flows generate and interact in SDDC.Applying the FUMA,the average compression rate in a flowtable can reach about 70 percent,and the average packet forwarding rate is about 98 percent.In a word,the MEM,an intelligent flowtable update mechanism MEM proposed in this thesis,can reduce the storage stress of flowtable storage effectively,and can improve the efficiency of packet forwarding through flow entry classification,prediction and update.