Research on congestion control mechanism based on active queue management

At present, TCP sliding window has been widely used on the Internet for end-to-end congestion control. However, with the increase of network users and the diversification of the business flows carried, end-to-end congestion control is facing severe challenges. Routers are the core entities of the Internet and the most direct sensor of network congestion status. Therefore, giving full play to the role of routers in congestion control is also an effective way to solve the congestion control problem. Moreover, routers may be more timely and even able to understand the status of the network in advance, and implement effective resource management strategies based on this to ensure that the network can effectively avoid congestion or recover from severe congestion as soon as possible.
The focus of this paper is queue management in router congestion control. Queue management controls the queue length by choosing when to discard which business flow packets.
1 Active queue management
1.1 Tail Drop
The most commonly used queue management strategy in routers is "tail drop". From the perspective of congestion control, it is a congestion recovery mechanism and also a passive queue management. However, it has three serious defects: continuous full queue status; deadlock of business flows on cache; global synchronization of business flows. To address these problems, "first drop" and "random drop" are proposed to be effective for deadlock and global synchronization, but they do not solve the problem of continuous full queues. In recent years, some scholars have proposed active queue management (AQM) [1-2] to allow routers to take some preventive measures before congestion occurs, so that the full queue problem can be solved more effectively.
The AQM strategy attempts to estimate the congestion at a certain point and mark it by dropping packets before the buffer is full. The corresponding congestion control strategy can reduce its transmission rate. This can avoid more serious congestion and try to reduce the packet loss rate and maintain a low average queue length. An AQM algorithm consists of two parts: judging the occurrence of congestion and deciding which packets to drop. Therefore, its performance depends on whether it is positive or conservative to judge the occurrence of congestion, and how to actively drop packets based on the judgment results.
1.2 RED
tail removal algorithm considers how to recover after congestion occurs. If measures are taken before congestion occurs, the full queue problem can be solved. AQM is designed to overcome the shortcomings of Tail Drop queues in the network. RED is the earliest AQM algorithm proposed.
The RED algorithm is based on the idea of ​​congestion avoidance. Instead of discarding the arriving packets after the cache is full, it uses the marking probability to discard some packets in advance to prevent congestion. At the same time, the RED algorithm does not adopt the source suppression strategy to immediately return the feedback information to the sender, but prompts the receiver by setting a flag bit, and then the receiver passes it to the sender; moreover, RED adjusts the packet loss rate through the average queue rather than the instant queue to absorb some short-term burst flows as much as possible. When the queue is frequently close to the full cache, the packet loss rate of RED is significantly lower than that of the tail loss queue.

Figure 1 shows the simulation curves of the two algorithms. From the results, we can conclude that as the traffic increases, both algorithms produce different degrees of delay, and under high load conditions, the tail removal algorithm exacerbates queue oscillation due to global synchronization [3-4].

2 Design Scheme
The RED algorithm and many improved algorithms cannot provide differentiated services for services with different QoS requirements, which cannot guarantee services with high QoS requirements. In this paper, an algorithm DS-RED (Different Serve RED) is proposed to achieve differentiated services. A dynamic threshold is set in the cache to control the packet loss rate, so that the cache can be dynamically allocated to each data flow. According to the different QoS requirements of each data flow, network resources can be dynamically adjusted to improve the utilization of network resources. By setting a threshold value and then dynamically adjusting the threshold value according to the loss of high and low priority packets, services with different QoS requirements can be treated differently, and the discard of high and low priority packets can reach a balance. Make fuller use of network resources.
2.1 Algorithm Design Goals
(1) Congestion avoidance and congestion control. Experiments show that to maintain high throughput and low latency in the network, congestion avoidance must be performed; as a remedy for the failure of congestion avoidance, congestion control must be implemented on the router to avoid congestion collapse in the network;
(2) Differentiated services for each data flow are achieved. A dynamic threshold is set in the cache to control the packet loss rate, so that the cache can be dynamically allocated to each data flow. Network resources can be dynamically adjusted according to the different Qos requirements of each data flow, thereby improving the utilization of network resources. It can set a threshold value and then dynamically adjust the threshold value according to the loss of high and low priority packets, so that services with different Qos requirements are treated differently, and the discard of high and low priority packets is balanced.
2.2 Algorithm idea
Set loss counters ch and cl for high and low priority data flows respectively, and each counter specifies a loss increment, such as kh and kl. When ch increases by kh, the threshold will be reduced by a certain value; and when cl increases by kl, the threshold will be reduced by a certain value. In this way, if too many high priority data packets are lost, the threshold value that increases the probability of discarding low priority data packets will be reduced to reduce the cache space of low priority data packets; conversely, if too many low priority data packets are lost, the threshold will be increased. The discard of high and low priority packets is balanced. Make network resources more fully utilized.
3 Performance comparison with RED algorithm
In order to compare the performance of RED and the new algorithm, a network simulation was performed. The network topology used in the simulation is shown in Figure 2.

S1 and S2 are data transmission sources, where S1 is assumed to send high-priority data streams, S2 is assumed to send low-priority data streams, and D1 and D2 are receiving ends. The bottleneck link is located between routers 1 and 2, and its link capacity is 64 kb/s. The capacity of the remaining links is 10 Mb/s. Each active queue management algorithm is set on router 1. The simulation time is 20 s, and the buffer size of the router is 50 packets. The parameter settings of the various algorithms introduced above are as follows: The basic parameters of RED are set to min_th=5, max_th=15, max_p=0.02. For the scheme DS-RED proposed in this paper, the parameters are set to min_th=5, max_th=15, max_p=0.2. The above simulation models are tested using the constant rate and file transfer FTP models [5-6], and the simulation results are shown in Figures 3 and 4 respectively.

From the simulation results, we can see that at a constant rate, the drop probability of the two algorithms is similar, but in the case of FTP file transfer, the drop probability of high-priority packets using the improved algorithm is significantly reduced, while the drop probability of low-priority packets is correspondingly increased. Before the improvement, the drop probability of high-priority packets was higher than that of low-priority packets, but after the improvement, the drop probability of high-priority packets was significantly lower than that of low-priority packets, and the improved algorithm also has significant improvements in time delay.
How to reduce the packet loss rate in the network, improve the utilization of the link, reduce the transmission delay, and prevent network collapse is very important for network research. When the data packet is lost before reaching the connection point or the link is idle for a long time or a large number of unnecessary data packets are retransmitted, it will cause a lot of waste of network resources. Therefore, the study of network congestion control becomes crucial.
Through MATLAB simulation, the performance of this method is compared with the RED algorithm. The results show that the improved congestion control mechanism can provide differentiated services and better meet Qos requirements.