Research On Key Technologies Of TCP Incast In Data Center Network

Data center is the core infrastructure of cloud computing technology.With the maturity and development of cloud computing technology,more and more complex business forms make the demand for data center network higher and higher.TCP(Transmission Control Protocol)protocol cannot be well applied to the current data center network.Because of its high bandwidth and low latency network environment,many-to-one transmission and communication mode makes TCP Incast problem,which leads to a sharp decline in network throughput and seriously affects the service performance of data center network.To solve this problem,the existing research mainly focuses on the sender(i.e.workstation),and the main means is to modify the workstation congestion window,so as to achieve the purpose of reducing congestion.There are two shortcomings in the existing technology: one is that the Incast problem has not been considered from the perspective of the whole network;the other is that the service time of the business cannot be fully guaranteed.In order to solve the above shortcomings,this paper proposes TCP protocol based on transmission rate allocation and scheduling algorithm based on guaranteed service time.(1)For the congestion problem caused by TCP Incast problem,in the TCP protocol based on transmission rate allocation,the traffic information of different TCP flows is collected by aggregating node,and the congestion window is adjusted for each service by global optimization method,which ensures the decrease of network congestion and improves the network throughput.In addition,in the process of allocating transmission rate,the protocol takes into account the difference of service priority and quality of service requirements,and gives priority to the needs of high priority services,while taking into account the business needs of low priority services.(2)In order to solve the problem of service transmission timeout and transmission failure caused by TCP Incast problem,in the scheduling algorithm based on guaranteeing service time,firstly,by adding access control mechanism in the sink node,the problem of more service transmission failure caused by simultaneous service for too many services is avoided.It improves the success rate of network transmission;secondly,by dividing priority queues for services with different urgency,priority is given to the quality of service of the service that will be timed out;finally,through scheduling algorithm,the service time of the whole service is reduced as far as possible in order to ensure that the service will not be timed out.