Research On Low Datency Data Center Networks Based On Transmission Control Mechanisms

With the rapid development of information technology,Internet and Internetbased application services have been integrated into every corner of our lives.Interactive applications such as Web search,social networking,advertising,recommendation systems and autopilot have high requirements for real-time performance.Data centers are the cornerstone of most Internet application operations,a fine data center network transmission performance is critical to improving overall service quality.Currently,the TCP protocol based on the packet loss mechanism is the most popular transmission protocol used in existing data center networks.Due to the data center has unique characteristics in terms of data traffic,network topology,and communication mode,it only supports the fair-sharing TCP protocol because it cannot effectively distinguish the priorities of data flow of different sizes and urgency,and ultimately leads to data flow.The average completion time is long,and the delaysensitive data flow misses its deadline.In view of the above problems,this paper optimizes the network transmission at the transport level for the flow level.The main research contents and contributions are as follows:In the private cloud and hybrid cloud data center scenario,we purposed a hybrid flow low-latency transmission scheme(DLSTCP)to define the priority of data flows according to the data flow urgency and data volume size information.The data center carries a variety of applications.For private clouds,the data flow characteristics can be divided into small flows with explicit deadlines and large flows that are not sensitive to time limits.The hybrid cloud will have additional small flows on the private cloud that are not explicitly notified of the deadline.When the network is congested,if the time-limited small flow is treated the same as the time-insensitive large flow,and the congestion window is fairly adjusted,the small flow will experience a long queue delay,and the existing transmission control protocol is also not correct.The urgency of the data flow is effectively differentiated,resulting in a higher level of data flow at the average completion time and the miss deadline.DLSTCP defines the priority of the data flow according to the tightness of the data flow and the size information of the data.When the network is congested,the congestion window of the data flow is dynamically adjusted according to the priority of the data flow and the degree of network congestion.This protocol is deployed in a data center in a distributed manner.It has been verified that DLSTCP can effectively improve the ratio of the average latency and the overall missed deadline.In the public cloud data center scenario,a low-latency transmission scheme(FIATCP)is proposed to define the priority of data flows according to the length o f time that the data flow has been sent and the amount of data that has been sent.Various types of data flows of different sizes and with or without explicit deadlines exist in the data center network.Most existing transmission schemes define the priority of data flows based on the information explicitly indicated by the size and deadline of the data flow.However,the types of tenants and applications supported by public cloud data centers are very complex.It is difficult to know the data flow size and time limit of each application in advance.To this end,FIATCP defines the priority of the data flow based on the length of time the data flow has been sent and the amount of data sent.When network congestion occurs,the algorithm dynamically adjusts the congestion window of each data flow according to the priority of the data flow.This protocol is deployed in a data center in a distributed manner,without any additional hardware requirements,and can coexist well with existing protocols.Experimental verification shows that FIATCP can effectively optimize the two indicators of low latency and missed time ratio without pre-acquiring any information of the data flow.