CN-122001877-A - QUIC stream concurrency number dynamic adjustment method and system for 5G/6G network file transmission

CN122001877ACN 122001877 ACN122001877 ACN 122001877ACN-122001877-A

Abstract

The invention discloses a QUIC stream concurrency number dynamic adjustment method and system for 5G/6G network file transmission, and belongs to the technical field of network communication. The method comprises the steps of establishing a rapid user datagram protocol internet connection and executing initial sampling based on a self-adaptive smooth round trip delay period, calculating utility values reflecting transmission comprehensive performance by utilizing collected link data, constructing a Gaussian process regression mapping model, calculating improvement potential of candidate concurrency numbers and selecting optimal target concurrency numbers based on the mapping model, synthesizing the number of dynamic adaptation flows of the optimal target concurrency numbers and the upper limit of the current concurrency numbers, executing capacity limiting processing and updating the upper limit of the concurrency numbers when link abnormal events are detected, and triggering a closed loop iteration optimization flow. The invention can obviously improve the convergence speed and throughput efficiency of batch file transmission of the 5G/6G cellular network.

Inventors

LI QIANG
Yang Wanghong
LI ZHENYA
QIN YIFANG
ZHOU XU

Assignees

中国科学院计算机网络信息中心

Dates

Publication Date: 20260508
Application Date: 20260316

Claims (10)

1. A method for dynamically adjusting the concurrency of QUIC streams for 5G/6G network file transfer, comprising the steps of: Establishing network connection and executing initial sampling to configure initial flow concurrency number, and acquiring link state data and service performance data in the transmission process; calculating utility values reflecting transmission performance based on the link state data and the service performance data, and constructing and updating a mapping model between the stream concurrency number and the utility values by utilizing the utility values and the corresponding stream concurrency numbers; calculating the improvement potential of the candidate concurrency number based on the mapping model, and selecting the candidate concurrency number with the maximum improvement potential as the optimal target concurrency number; Determining the actually executed concurrency number according to the optimal target concurrency number and the upper limit of the concurrency number of the current link, and realizing the rapid adaptation of the concurrency number by adjusting the number of the concurrency streams; And monitoring the real-time link performance under the actual concurrent number, dynamically adjusting the sampling period self-adaptive adjustment coefficient according to the link fluctuation state, and updating the upper limit of the concurrent number when the abnormal event of the base station occurs to the link so as to trigger the concurrent number adaptation flow of the next round.
2. The method of claim 1, wherein establishing a network connection and performing initial sampling to configure an initial number of stream concurrency comprises: acquiring the total number of files to be transmitted, and initializing protocol basic parameters, wherein the protocol basic parameters comprise the maximum flow number, a congestion control algorithm and connection timeout time; initializing Bayesian optimization model parameters, and configuring an observation record mechanism consisting of an observation set and a sliding window; Configuring a multidimensional parameter acquisition module, a Bayesian optimization modeling module and a dynamic adjustment module, and selecting a sample value from a preset discrete concurrency number set as an initial stream concurrency number; setting an initial value of a sampling period self-adaptive adjustment coefficient, defining a value interval, and configuring statistical variables for recording abnormal states of a base station, an upper limit of concurrent numbers and the sampling period.
3. The method of claim 1, wherein obtaining link state data and traffic performance data during transmission comprises: Taking the product of the self-adaptive adjustment coefficient of the sampling period and the smooth round trip delay as the sampling period, and collecting the concurrent number of the flow in the current period, the current time stamp, the accumulated received data packet number and the maximum packet sequence number received in the period in real time; And acquiring the original round trip delay of the current period, and updating the smooth round trip delay reflecting the long-term stable state of the link according to the original round trip delay.
4. The method of claim 3, wherein updating the smoothed round trip delay comprises: judging whether the current sampling is the first sampling or not, if so, taking the original round trip delay as the smooth round trip delay; and if the sampling is the subsequent sampling, carrying out weighted average calculation on the smooth round-trip delay of the previous period and the original round-trip delay of the current period according to a preset smoothing factor to obtain updated smooth round-trip delay.
5. The method of claim 1, wherein calculating a utility value reflecting transmission performance comprises: Calculating the difference value of the accumulated received byte number of the current period and the last period to obtain the period received byte number, and calculating the period duration according to the time stamp; Dividing the number of the periodically received bytes by the product of the period duration and the stream concurrency number, and calculating to obtain single-stream average throughput; The maximum packet sequence number received in the period is differed from the maximum packet sequence number of the previous period to obtain the total number of estimated transmitted packets, and the accumulated received data packet number is differed from the accumulated received data packet number of the previous period to determine the total number of actual received packets; Subtracting the total number of the actual received packets from the total number of the estimated transmitted packets, taking a non-negative value to obtain an estimated packet loss number, and dividing the estimated packet loss number by the total number of the estimated transmitted packets to obtain a packet loss rate; and combining a preset concurrent overload penalty coefficient and a packet loss penalty coefficient, and calculating single-flow average throughput, packet loss rate and flow concurrency number through a utility function to obtain a utility value reflecting the comprehensive transmission performance.
6. The method of claim 5, wherein calculating the single-flow average throughput, the packet loss rate, and the number of flows concurrency by the utility function to obtain the utility value reflecting the overall performance of the transmission comprises: calculating a logarithmic gain term according to the single-stream average throughput, and multiplying the stream concurrency number, the single-stream average throughput and the logarithmic gain term to obtain an original performance gain; calculating an overload suppression item according to the stream concurrency number and the concurrency overload penalty coefficient, and weakening the original performance gain by using the overload suppression item to obtain a primary utility value; Calculating the product of the stream concurrency number, the single stream average throughput, the packet loss rate and the packet loss penalty coefficient to obtain a packet loss penalty item; And subtracting the packet loss penalty term from the preliminary utility value to obtain a final utility value.
7. The method of claim 1, wherein dynamically adjusting the sampling period adaptation coefficients comprises: calculating a smooth round trip delay difference value, a smooth round trip delay ratio and a total throughput ratio of a current period and a previous period; Determining a link fluctuation level based on the smooth round trip delay ratio, the smooth round trip delay difference and the total throughput ratio, wherein the fluctuation level comprises slight fluctuation, medium fluctuation and serious fluctuation; If the fluctuation is judged to occur, carrying out reduction updating on the sampling period self-adaptive adjustment coefficient according to the preset shrinkage proportion corresponding to the fluctuation level, restraining the updated coefficient in a preset value interval, and setting an adjustment cooling period of the preset period; If the link is in a stable state and the smooth round trip delay is smaller than a preset delay threshold, the sampling period self-adaptive adjustment coefficient is increased round by round according to a preset step length, and if the link is in other stable states, the step increase is executed according to a preset longer period until the preset upper limit value is reached.
8. The method of claim 1, wherein determining that a link has a base station anomaly event comprises: If the current period smooth round trip delay ratio, the smooth round trip delay difference absolute value and the total throughput ratio respectively reach the corresponding preset abnormal thresholds, judging that the basic abnormal condition is met; Counting the hit times of the basic abnormal conditions through a sliding window with a preset length, and judging that the common base station is abnormal if the hit times reach a preset frequency threshold; If the smooth round trip delay ratio, the smooth round trip delay difference, the total throughput ratio or the packet loss rate increment meet the preset serious abnormal triggering condition, the base station is directly judged to be abnormal.
9. The method of claim 8, wherein updating the upper limit on the concurrency limit upon determining that a link has a base station exception, comprises: Multiplying the current cycle flow concurrency number by the total throughput ratio to obtain a preliminary upper limit value, and selecting the minimum value of the preliminary upper limit value and the historical upper limit value as the current concurrency number upper limit; Setting an abnormal limit cooling period of a preset round, and obtaining the minimum value of the optimal target concurrency number and the upper limit of the current concurrency number by the actually executed concurrency number in the cooling period; After the cooling period is finished, if the total throughput is restored to the specified proportion before abnormality, the upper limit of the stepwise rising concurrency number is executed.
10. A qic stream concurrency adjustment system for 5G/6G network file transfer, comprising: The system comprises a multi-dimensional parameter acquisition module, a sampling period self-adaptive adjustment coefficient, a base station abnormality judgment and concurrency number upper limit control module and a base station abnormality judgment module, wherein the multi-dimensional parameter acquisition module is used for establishing network connection, configuring initial concurrency number and acquiring real-time link state data and business performance data; The Bayesian optimization modeling module is used for constructing and updating a mapping model between the stream concurrency number and the utility value by utilizing the utility value and the corresponding stream concurrency number, calculating the improvement potential of the candidate concurrency number based on the mapping model, and selecting the candidate concurrency number with the maximum improvement potential as the optimal target concurrency number; the dynamic adjustment module is used for determining the actual execution concurrency number by combining the optimal target concurrency number and the upper limit of the concurrency number of the current link, and realizing the rapid adaptation of the actual execution concurrency number by dynamically adjusting the number of the concurrency streams; The multi-dimensional parameter acquisition module captures link fluctuation through self-adaptive adjustment coefficients of a self-adaptive scaling sampling period, and cooperates with the dynamic adjustment module to complete capacity limiting and recovery under an abnormal scene.

Description

QUIC stream concurrency number dynamic adjustment method and system for 5G/6G network file transmission Technical Field The invention belongs to the technical field of network communication, and particularly relates to a QUIC (quick flow response) stream concurrency number dynamic adjustment method and system for 5G/6G network file transmission. Background With the large-scale application of the 5G technology and the evolution iteration of the 6G technology, the 5G/6G cellular network becomes one of core bearing networks for mass file transmission by virtue of the characteristics of enhancing mobile broadband, ultra-high reliability, low-delay communication and the like. The 6G network further integrates space-earth integrated networking, millimeter wave/terahertz and other communication technologies, so that the transmission rate of the terahertz bit level and the microsecond time delay are realized, and the complexity of dynamic fluctuation of the link state and link switching is further increased. In the network transmission protocol field, the traditional TCP (transmission control protocol) takes the dominant role for a long time, but has a plurality of inherent defects in the high dynamic and high concurrency scene (1) the TCP adopts three-way handshake to establish connection, the handshake process is too long in long-distance communication or high-time delay network, the connection establishment delay is obviously increased, (2) the subsequent data transmission of the TCP needs to depend on a complex serial number and retransmission mechanism, which can ensure the data transmission reliability, but generates huge protocol cost and occupies additional network and system resources, (3) the TCP is based on byte stream transmission, even in the HTTP/2 protocol, multiplexing is still limited by the transmission characteristic of the TCP, the problem of queue head blocking exists, the packet loss of a single data stream can lead to the stagnation of all the concurrency data streams, (4) the concurrent transmission of the TCP depends on a multi-connection mechanism, each connection is independently controlled by congestion, the bandwidth allocation is easy to be unbalanced, the maintenance of the multi-connection further increases the system resource cost, (5) the safe transmission of the TCP needs to depend on an additional TLS/layer, and the subsequent handshake and encryption further extend the transmission delay. To address the above-mentioned inherent deficiencies of the TCP protocol, the QUIC (Quick UDP Internet Connection) protocol has evolved. The QUIC protocol is realized based on UDP, so that the defects of TCP are effectively overcome, and the method has the core advantages that (1) Connection establishment of 0-RTT or 1-RTT is supported, connection delay is greatly shortened, session multiplexing is supported, reconnection efficiency is further improved, (2) a TLS1.3 encryption function is originally integrated, a security layer is not required to be additionally overlapped, transmission cost is reduced, (3) a stream-level multiplexing mechanism is adopted, packet loss of a single stream cannot influence normal transmission of other streams, the problem of queue head blocking in batch data transmission is avoided, (4) a Connection state is maintained through a Connection identifier (Connection ID), seamless Connection migration is supported, network switching under a mobile scene of an adaptive terminal is adapted, and (5) a congestion control algorithm can be plugged and plugged, and flexible switching can be realized according to network scenes, so that the dynamic characteristics of a 5G/6G network are adapted. However, neither TCP nor qic can fully adapt to the high-speed changes of 5G networks. The fixed uplink/downlink ratio of the 5G/6G access network causes a tidal effect on the load, and the performance of the cellular network can be significantly changed due to factors such as wireless interference, signal shielding, base station switching and the like, especially in a mobile environment. The existing method generally lacks the accurate determination capability and the quick response mechanism aiming at the index change reason in the 5G/6G link transmission, so that the real-time accurate matching of the sending rate and the link state is difficult to realize in single-connection single-stream transmission, the channel congestion and even packet loss are caused in the case of superfluous transmission, the channel utilization rate is lost in the case of less occurrence, and the bandwidth allocation fairness of the cellular access network is easily damaged in the case of multi-connection concurrent transmission. The multiplexing characteristic of the QUIC protocol supports the creation of multiple concurrent flows in a single connection to improve transmission throughput, compared with the concurrency of multiple connections, the concurrency cost of the single connection and the multiple