Research On Memory Optimization Technology Based On SDN

With the rapid development of Internet technology,the number of users and network traffic have exploded.Remote Direct Memory Access(RDMA)stands out by virtue of its low latency,low CPU usage,and high throughput.RDMA over Converged Ethernet(Ro CE)mainly relies on mechanisms such as Priority-based Flow Control(PFC)to guarantee lossless packet loss.However,in the actual data center network operation,these mechanisms may cause circular buffer dependency between the receiving queues in the switch memory created due to data transmission,which will cause the serious consequence of permanent blocking of PFC memory.At the same time,traditional network architectures have the disadvantages of low flexibility and low management and control capabilities,while Software Defined Network(SDN)architectures have powerful functions such as numerical control separation,programmability,and centralized management and control,enabling the network to be flexibly and dynamically expanded and configured.To adapt to new business and new needs.Therefore,this paper introduces the SDN architecture into the construction of a data center network carried by RDMA,and proposes a SDN-based data center switch memory blocking optimization strategy.Through centralized control of the network,it solves the complexity and difficulty of the traditional data center network architecture.Regulatory issues.This optimization strategy aims to avoid the permanent blocking of memory in a manner that dynamically senses and breaks the circular buffer dependency.While achieving the purpose of optimizing the switch memory in the network,it guarantees the efficient and stable operation of the RDMA data center network.The thesis focuses on this focus,and the main contents include the following aspects.First,the related technologies of RDMA and SDN are explored.Through experimental and theoretical research,the in-depth analysis of the process and the formation of permanent memory blockage in Ro CE networks is mainly conducted from the aspects of multiple data flow states,limit rate states,and asynchronous update of ARP entries.Reasons,summarize and discuss multi-angle solutions,and analyze the SDN architecture features,Open Flow protocol,and SDN controller in detail.Based on this,the consideration angle of this article is clarified.Secondly,by abstracting the queuing process in the switch memory in detail,an improved memory blocking balance model is constructed.The probability of ignoring high-priority queue packets(ignoring factor ?)is defined in combination with memory blocking characteristics,which can temporarily weaken the priority of high-priority queues.From the perspective of reducing the probability of memory blocking,solve the problem of including the ignoring factor Priority blocking balance model.Matlab proves its effectiveness,and NS-3 simulations show that the stress relief capability of low-priority receive queues in memory can be increased by about 6% when the ignore factor is enabled.Finally,an SDN-based data center switch memory blocking optimization strategy is proposed.By adding switch memory status monitoring submodules,path abnormality sensing submodules,forwarding control decision deployment submodules,and alternative path triggers to the SDN controller,To achieve the improvement and supplement of the original PFC mechanism,from the perspective of eliminating the circular buffer dependency,specific optimization measures for different objects are given,and the problem of permanent memory blockage caused by the original mechanism to the data center network is solved.Through NS-3 design simulation experiments,it is verified that the SDN-based data center switch memory permanent blocking optimization strategy proposed in the paper can effectively and reasonably optimize the memory of the data center switch,avoid permanent blocking of PFC memory,and ensure the smooth and efficient operation of the network.