With the rapid growth in the number of low-Earth orbit (LEO) satellites, especially with the increasing application of satellite optical communication networks, efficiently and flexibly managing and optimizing the communication links and network traffic between satellites has become an urgent problem. Traditional satellite network routing and load balancing methods often perform poorly in the face of high dynamics, complex topologies, and unbalanced satellite-ground traffic distribution, which leads to issues such as link congestion, traffic bottlenecks, and resource utilization inequality. This reduces satellite resource utilization and causes network load imbalance. Therefore, it is particularly important to study an adaptive load balancing algorithm that can effectively handle network load fluctuations, reduce the blocking rate, and improve throughput.

We address the problem from the perspective of satellite-ground traffic balancing, considering the onboard and satellite-ground traffic distribution characteristics of LEO satellite optical networks. Using the polar Walker constellation model, a LEO satellite optical network scenario is constructed, and two load balancing algorithms are proposed: the redundancy-aware load balancing (RALB) algorithm and the detour multipath routing algorithm for satellite-ground traffic (DMRA-SGT). First, the RALB algorithm optimizes the Dijkstra path search logic by utilizing multiple equivalent paths in inter-satellite links. It removes overloaded links, prioritizes lightly loaded ones, and effectively alleviates network bottlenecks, thereby improving resource utilization efficiency. Second, the DMRA-SGT algorithm combines the congestion distribution index and uses traffic packet transmission to redirect some traffic from the default route to detour paths located farther from the core satellite region, which loads are lighter. The routing tables for both detour paths and shortest paths are updated synchronously to avoid temporary loops. To quickly respond to data forwarding and congestion paths, a fast distributed routing protocol (FDR) is designed as the default routing scheme for the satellite-ground traffic detour algorithm. This protocol requires fewer onboard resources for calculating the shortest and long-distance detour paths, thus achieving effective traffic distribution and mitigating cascading congestion.

On MATLAB R2018b and satellite simulation software STK 11.6, we build a polar-orbiting satellite constellation, including 288 satellites (constellation parameters: 288/12/24, 1400 km) and 16 ground stations, and use OPNET Modeler 14.5 for simulation verification. To measure the degree of network congestion under different algorithms, we use the traffic blocking ratio (TBR) as an indicator. Simulation results show that the DMRA-SGT algorithm performs the best among the four algorithms, while the Dijkstra algorithm performs the worst. When the total traffic volume is lower than 5.7 Tbit/min, the TBR of Dijkstra’s algorithm is zero. However, when the traffic volume exceeds this threshold, the network becomes congested, and the TBR increases rapidly. When the system reaches the maximum of 9.9 Tbit/min, the TBR reaches 26.28%. This is because Dijkstra’s algorithm needs to store the entire network information, resulting in excessive memory consumption in complex networks. In contrast, the RALB algorithm begins to show congestion when the total traffic volume reaches 6 Tbit/min and the TBR peaks of 21.71% at 9.9 Tbit/min. Although the RALB algorithm considers link loads to avoid heavily loaded links, its effect is limited. The SGC-LB algorithm works by directing congestion traffic to the nearest gateway, but as the link load is not taken into account, the network begins to experience congestion at 6.6 Tbit/min, with the final TBR peak of 2.08%. However, the congestion in the DMRA-SGT algorithm only appears when the total traffic volume reaches 6.6 Tbit/min, and when it reaches 9.9 Tbit/min, the peak value of TBR is only 0.49%. This is 25.79 percentage points, 21.22 percentage points, and 1.59 percentage points lower than those of Dijkstra, RALB, and SGC-LB, respectively. Regarding the evaluation of the average packet loss rate and total throughput, the simulation results show that the DMRA-SGT algorithm performs the best in terms of packet loss rate. Especially when the total traffic is 10 Tbit/min, the packet loss rate of DMRA-SGT is 3.17%, which is much lower than that of the other algorithms. The reason for this is that DMRA-SGT reduces congestion and packet loss by bypassing routes to avoid high-traffic areas. SGC-LB and RALB also reduce the packet loss rate by choosing lower load paths, although not as effectively as DMRA-SGT. The Dijkstra algorithm, however, selects the congested links, which results in the highest packet loss rate, reaching 14.12%. In terms of total throughput, the DMRA-SGT algorithm also demonstrates the best performance. When the total traffic is 9.9 Tbit/min, the total throughput of DMRA-SGT is 9821.6 Gbit, which is 18.2% higher than that of Dijkstra’s algorithm. Although the increase in total throughput is not as significant as that of the blocking rate and packet loss rate, it still highlights the advantages of DMRA-SGT under high loads. In summary, the DMRA-SGT algorithm outperforms the Dijkstra, RALB, and SGC-LB algorithms in terms of traffic blocking rate, packet loss rate, and total throughput, which demonstrates superior network performance.

In this paper, we consider the effect of traffic load on satellite optical networks and propose a satellite-based redundancy-aware load balancing algorithm. This algorithm utilizes on-board resource redundancy to reduce network congestion and improve network throughput. We then introduce the DMRA-SGT algorithm, which combines the shortest path with long-range detour paths. Based on network state information, the algorithm computes a congestion distribution index and selects long detour paths using a detour multipath algorithm according to a fast distributed routing protocol. Through simulation, we compare and analyze the performance of the DMRA-SGT algorithm under high-throughput traffic, which shows that it outperforms other algorithms in terms of traffic blocking rate, packet loss rate, and total throughput. Specifically, compared to the widely used Dijkstra algorithm, the blocking rate is reduced by 25.79 percentage points, the packet loss rate is reduced by 10.95 percentage points, and there is a significant improvement in total throughput.