Research On The Resource Optimization Of Dataplane In Software Defined Interconnection

New generation network architectures such as Software-Defined Interconnection(SDI)separate the network controling and data forwarding functions and provide the open programmable interfaces,breaking down the existing integrated device barriers,enhancing the network service capabilities,improving the flexibility and extendibility,and powerfully promoting innovative deployment of network functions and applications.The dataplane serves as the basic support for control forwarding and data processing,and provides intrinsic impetus to development and evolution for new networks.The network elements in dataplane parse data and extract fields of the data flow according to the protocol parsing graph,and simultaneously match fields and process data packet according to the flow table delivered by the controllers.Therefore,the dataplane determines the processing speed and service capability of the entire network,and requires to achieve flexible and efficient network functions under the limited resources.At present,there are still some problems that need to be resolved in dataplane.It is difficult to achieve a balance between flexibility and high efficiency in the parsing state transition structure.The multiple flow table achieves resource optimization,at the same time it brings more matching lookup cycles.The expansion of the table size in the network,combined with the structure of multi-field matching field,results in more redundancy between existing entries and low storage efficiency.Supported by the National Science Technology Major Project(No.2016ZX01012101)--“The xxx device based on software-defined interconnection”,this dissertation takes the softwaredefined thought as theoretical guide to analyze the existing parsing entries and matching entries in dataplane.Three aspects of the generation of programmable parsing entries,the storage construction of multiple flow tables and the redundant information in storage of different entries are studied and optimized.The purpose is to improve the resource utilization efficiency and data processing performance of the dataplane in SDI,and then establish a flexible and efficient message switch platform.The main contents of the dissertation are as follows:1.Aiming at the problem that the parsers of current network forwarding devices are difficult to deal with the emerging network requirements,this paper proposes a Programmable parse Table Generation System based dynamic programming(PTGS),using programmable parsing table structure.First of all,by configuring different protocol parsing graphs,the protocol can be flexibly customized.In addition,the corresponding dynamic programming algorithm aggregates short-byte headers,thus parsing multiple packet headers in single cycle.Finally,simulation experiments show that the solution reduces parse cycles under lower resource utilization.2.Aiming at the balance between the table series and the matching delay caused by the partitioning process of multiple flow table,a Category-first method for Multiple Table Establishment is proposed(CMTE).By analyzing the correlation between matching fields in different entry categories,the matching field compatibility is defined,and an optimization model based on maximum compatibility is established.In addition,a category-first algorithm for solving multiple flow table partition set is given.Then the applicable matching lookup structure and the corresponding entry update strategy are proposed for this scheme.Finally,through simulation experiments,it is verified that this scheme effectively mitigates the increasing of matching lookup cycles brought by multiple flow table under the improvement of resource utilization.3.Aiming at the problem of continuous increasing in bit-width and the scale expansion of entries caused by the rapid development of network,a compression method based on Bit Extraction of Independent rule Subsets for packet classification(BEIS)is proposed.First of all,some fields are merged based on the redundancy information by analyzing the logical relationships among fields in multiple field tables.Secondly,based on the independence of rules,the packet classification ruleset is divided into several independent rule subsets.Then,distinguishable bit extraction is performed on independent rule subsets to further compress the storage space.Finally,simulation experiments and comparative analysis show that the proposed scheme can effectively reduce the TCAM storage space.