A Distributed System Design Based on DDS Middleware for Dynamic Real-Time Load Balancing

A distributed system is a computing environment, where a set of nodes are separated among geographically distributed locations. The distributed nodes can work together cooperatively to solve one big task and appear as a single cohesive system to the end user. However, distributing the workload among nodes evenly is crucial to avoid bottlenecks, where certain nodes become overloaded while others remain underutilized. Therefore, a dynamic load balancing algorithm is important to continuously monitor the system and adjust the allocation of tasks in real-time. This requires dynamic real-time updates from the distributed nodes to the controller of the distributed system. However, knowing the status of each node is a challenging task because of the delay that will be introduced by sending its status via the network. Moreover, the nodes in a distributed system are of different types and architectures. Hence, the transmission of nodes' status is a challenge due to their incompatibility. Therefore, we present a distributed system design based on data distribution service (DDS) with a fault-tolerant dual-controller architecture to manage real-time data transmission and enable dynamic load balancing across distributed nodes. Four experiments were conducted to evaluate the effectiveness of the proposed design. DDS achieved lower latency (max 250 ms vs. 400 ms) and consistent load balance (uniformity score 2.83 vs. 2.59) compared to Zenoh in real-world scenarios. Ultimately, DDS is the more reliable option for systems requiring precise, low-latency decision-making in load balancing, whereas Zenoh is better suited for high-throughput, dynamic, and less deterministic environments.