CN-122027563-A - Low-invasive congestion control method based on RTT and AoI fusion awareness

CN122027563ACN 122027563 ACN122027563 ACN 122027563ACN-122027563-A

Abstract

The invention discloses a low-invasiveness congestion control method based on fusion awareness of RTT and AoI, belonging to the technical field of computer network congestion control; the method comprises the steps of generating a time stamp and calculating information age by the time stamp received by a receiving end, fusing normalized historical minimum round trip delay with the weight of the historical minimum information age to generate a comprehensive congestion index, comparing the comprehensive congestion index with a historical reference index and a previous measurement period index to determine a network state, and selecting a window increasing or window decreasing strategy according to the network state. The invention is sensitive to forward path congestion through AoI, avoids misjudgment caused by reverse path interference, enables an algorithm to recognize real congestion state earlier and more accurately in a complex network environment, can clearly distinguish different network phases such as congestion worsening, relieving, idle and the like, and drives differential growth and reduction strategies, thereby realizing dynamic optimization among low invasiveness, high utilization rate and quick response.

Inventors

GUO YONGAN
LIANG TIANYU
AN XIN
SHE HAO
LUO BINGQING

Assignees

南京邮电大学

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (7)

1. A low-invasive congestion control method based on fusion awareness of RTT and AoI is characterized by comprising the following steps: step S1, acquiring round trip delay, a data packet generation time stamp and a receiving time stamp of the data packet generation time stamp at a receiving end based on confirmation feedback to calculate information age; step S2, updating the historical minimum round trip delay and the historical minimum information age as reference values, and carrying out normalization processing; step S3, fusing the normalized historical minimum round trip delay and the weight of the historical minimum information age to generate a comprehensive congestion index; s4, comparing the comprehensive congestion index with a historical reference index and a previous measurement period index to determine a network state, and updating the historical reference index and the previous measurement period index according to a preset rule; Step S5, selecting a window increase or window decrease strategy according to the network state; under the non-congestion state, determining a target sending rate according to a preset growth function and converting the target sending rate to obtain a target congestion window; in the congestion state, adopting a reduction factor corresponding to the network state to reduce the congestion window; And S6, controlling data transmission according to the updated congestion window.
2. The method for controlling low-intrusive congestion based on fusion awareness of RTT and AoI as set forth in claim 1, wherein in step S1, the round trip delay, the packet generation timestamp and the reception timestamp at the receiving end are obtained based on acknowledgement feedback to calculate the information age, specifically: the sending end generates a time stamp T_generation for each data packet, the receiving end carries a receiving time stamp T_receiver in a returned acknowledgement packet ACK, and an expression for calculating information age is as follows: 。
3. the method for controlling low-intrusive congestion based on fusion awareness of RTT and AoI as set forth in claim 1, wherein in the step S2, the historical minimum round trip delay and the historical minimum information age are updated as reference values, and normalization processing is performed, specifically: The updated formula of the historical minimum round trip delay is: ; Wherein, the Representing the historical minimum round trip delay previously stored, Representing a round trip delay sampling value obtained by current measurement; the update formula of the historical minimum information age is: ; Wherein, the Representing the age of the currently stored historical minimum information, Representing the current calculated information age sampling value; the updating strategy is a conservative updating strategy, namely, the updating is carried out only when the current sampling value is smaller than the historical minimum value, and the updated value is the lowest value And (3) with Respectively as the reference values of RTT and AoI in the normalization process.
4. The method for controlling low-intrusive congestion based on fusion awareness of RTT and AoI as claimed in claim 1, wherein in the step S3, the normalized historical minimum round trip delay and the weight of the historical minimum information age are fused to generate a comprehensive congestion index, specifically: the comprehensive congestion index expression is: ; wherein RTT is the current measured round trip delay, min_RTT is the minimum RTT observed in the connection history, which represents the base delay without queuing, and min_ AoI is the minimum observed in the connection history Representing the best one-way delay; when each ACK arrives, calculating the current comprehensive index value according to the current sampling value : ; Wherein, RTT_sample is the current RTT sampling value, aoI _sample is the current AoI sampling value; the weight of the RTT item represents the sensitivity degree of the method to RTT change; Weight of AoI, represents the sensitivity of the method to AoI changes, where 。
5. The method for low-intrusive congestion control based on fusion awareness of RTT and AoI as set forth in claim 1, wherein in step S4, the integrated congestion indicator is compared with the historical reference indicator and the previous measurement period indicator to determine the network status, and the historical reference indicator and the previous measurement period indicator are updated according to a predetermined rule, which is specifically as follows: describing network dynamics by adopting a six-state finite state machine, and comparing the comprehensive index C twice; representing the current comprehensive index value calculated according to the sampling value; the table records the minimum C value since the connection began; A C value representing the last measurement period; the state determination rule is as follows: If it is And is also provided with Judging that the congestion is worsened in the state I; If it is And is also provided with Judging that the congestion is stable in the state II; If it is And is also provided with Judging that the congestion is relieved in the state III; If it is And is also provided with Judging that the congestion is relieved in the state IV; If it is And is also provided with Judging that the network is idle; If it is And is also provided with Judging that the state VI is an ideal state; the reference index update rule is as follows: When (when) In the time-course of which the first and second contact surfaces, ← ; The value of C, representing the last measurement period, after each measurement period, ← ; History reference value And The update of (1) adopts a 'small-case update' strategy: ; 。
6. The method for controlling low-intrusive congestion based on fusion awareness of RTT and AoI as set forth in claim 1, wherein in the step S5, in a non-congestion state, the target sending rate is determined according to a preset growth function and converted into a target congestion window, specifically: The preset growth function is a composite growth function, and the target sending rate The calculation formula of (2) is as follows: ; Wherein, the Indicating the target transmission rate at which the data is to be transmitted, Representing the difference between the current time and the current window-growing period start time, And For a growth factor that is determined by the current network state, a smaller value indicates a more aggressive growth, Representing a transmission rate at the beginning of a current window growth period; Scaling a target sending rate to a target congestion window The formula of (2) is: ; Wherein, the Representing reference delay estimation, wherein the reference delay estimation is determined by the minimum historical round trip delay and the minimum historical information age together and is consistent with a reference used in the calculation of the comprehensive index C; the non-congestion state corresponds to state III, IV, V or VI in the state machine.
7. The method for low-intrusive congestion control based on fusion awareness of RTT and AoI according to claim 1, wherein in the congestion state in step S5, a reduction factor corresponding to the network state is adopted to reduce the congestion window, specifically: Reducing congestion windows, and reducing congestion windows The calculation formula of (2) is as follows: ; Wherein, the Indicating the congestion window size before the reduction, Representing a reduction factor determined by the current network state, with a value range of 0 < < 1, Representing a maximum operation to ensure that the congestion window is not less than 2 segments; the reduction factor Setting according to the difference of congestion severity, when the network state is state I, setting Smaller, when the network state is state II, setting Larger, the congestion state corresponds to state I or state II in the state machine.

Description

Low-invasive congestion control method based on RTT and AoI fusion awareness Technical Field The invention belongs to the technical field of computer network congestion control, and particularly relates to a low-invasive congestion control method based on fusion awareness of RTT and AoI. Background With the explosive growth of business such as industrial internet of things, real-time audio and video communication, cloud computing and the like, modern networks have become a complex hybrid system bearing multiple priority and multiple service quality requirements. In this context, low priority (Less than Best Effort, LBE) congestion control techniques have been developed, and the core goal is to design a polite data transmission behavior, so that background traffic such as file backup, software update, etc. can silently make full use of network idle capacity, and once high priority traffic such as video conferences, remote control, etc. is detected to enter competition, bandwidth can be quickly and actively deferred, thereby improving the overall network resource utilization rate while ensuring the key service experience. Although this concept is clear, the prior art solution exposes several deep and systematic drawbacks in practical application, making it difficult to stably achieve the design objective in a dynamic and complex real network environment. The limitations of the existing schemes are primarily rooted in their single and interference-prone perception mechanism. Mainstream methods generally rely on a certain delay metric (such as one-way delay or round trip delay) as the only basis for determining network congestion. Such single-dimensional observations present inherent vulnerability, for example, algorithms that rely on one-way delay are severely limited by the accuracy of end-to-end clock synchronization, which is highly prone to false positives in wireless or wide area networks, while algorithms that rely on round-trip delay fail to distinguish whether the increase in delay is due to queue pile-up of the forward data path or to congestion of the reverse acknowledgment path or processing delay of the receiving end host. This causes a common problem in that when a reverse path is in condition, the transmitting end is forced to reduce the transmission rate, resulting in waste of forward bandwidth. Although research attempts to improve perceived robustness through statistical trend analysis have not broken through the constraint of a single signal source per se, perceived accuracy and timeliness still face serious challenges in environments where path characteristics are abrupt or continuously congested. Second, existing schemes exhibit significant stiffness in the adaptability of the control logic. To ensure low invasiveness, many algorithms employ extremely conservative and fixed growth strategies, which make them unresponsive when the network suddenly experiences a large amount of free bandwidth, resulting in long-term low link utilization. On the other hand, when signs of congestion are detected, their response strategies tend to lack fine granularity discrimination. Whether the deceleration is triggered based on a fixed threshold or adjusted towards a static target delay, this control scheme is difficult to scale, differential response depending on the severity and trend of the congestion. The result is often either excessive back-off in light congestion, sacrificing efficiency, or insufficient back-off or hysteresis in congestion aggravating, failing to effectively alleviate queuing, thus affecting the core design goals-achieving a dynamic, intelligent balance between high utilization and low invasiveness. Finally, parameter sensitivity and fairness dilemma are engineering practice challenges that hamper large-scale deployment of existing solutions. To implement the design function, these algorithms inevitably introduce a number of control parameters (e.g., growth factors, reduction factors, various types of detection thresholds). The optimal values of these parameters are highly dependent on the particular network environment (e.g., bandwidth, baseline latency, buffer size). In dynamically changing networks or heterogeneous mixed traffic scenarios, it is difficult for a fixed parameter set to always maintain optimal performance, resulting in unstable algorithm behavior and even performance concurrency. More prominent is the fairness problem, especially between the same type of LBE flows. A typical drawback is that newly added data flows may gain an unfair competitive advantage (e.g., mistaking the queue delay caused by existing flows into its own baseline and thus performing more aggressive) due to deviations or couplings in the estimation of the network "baseline state", which is contrary to the principle that the remaining bandwidth should be shared fairly between LBE flows. Solving these problems often requires further complexity in deployment and operation by complex parameter tuning or