Abstract
AUTOSAR (Automotive Open System Architecture) has become the global standard for automotive electronic software development, improving portability and development efficiency. However, the deployment of thousands of software components (SWCs) across Electronic Control Units (ECUs) presents a significant optimization challenge. This paper introduces the Traffic-Degree Clustering Algorithm (TDCA), which optimizes deployment by prioritizing components based on their communication "degrees" and data traffic. Experimental results show that TDCA significantly reduces bus load and improves ECU load balancing compared to traditional heuristic and linear optimization methods.
Introduction
Modern vehicles are increasingly defined by their software, running tens of thousands of components across dozens of ECUs. To manage this complexity, the industry relies on the layered AUTOSAR framework.
A critical bottleneck in this architecture is deployment: deciding which SWC resides on which ECU. Inefficient mapping leads to congested communication buses and unbalanced ECU utilization, which can compromise system reliability. This is an NP-hard optimization problem that previous approaches, such as the traffic-based CAT algorithm, often fail to solve optimally due to reliance on random initial placements.
Methodology
The proposed TDCA algorithm and its variant, TDSCA (Traffic-Degree-Subset-Cluster-Algorithm), introduce a structured approach to selecting "cluster centers" for deployment based on two primary metrics:
- Degree Distribution ($D_i$): Instead of random selection, TDCA calculates the "degree" of each SWC—the number of other components it communicates with. The algorithm selects the $k$ components with the highest degrees as initial anchors for the $k$ available ECUs.
- Communication Traffic ($ct$): Remaining SWCs are mapped to ECUs based on the volume of data exchanged. By keeping high-traffic pairs on the same ECU, the algorithm "internalizes" communication, keeping it off the shared vehicle bus.
To ensure physical schedulability, TDCA utilizes a Dynamic Fitness Function based on Rate Monotonic (RM) scheduling theory, targeting a utilization threshold of 0.69. A penalty function dynamically balances the trade-off between reducing bus traffic and equalizing ECU processing loads.
Experimental Evaluation
The researchers evaluated TDCA using three distinct data models:
- AIS Dataset: 904 atomic SWCs extracted from real-world AUTOSAR application examples.
- PLP (Planted l-partition) Model: Synthetic connected graphs representing various software architectures.
- RGE (Random Graph Exponential) Model: Synthetic models mimicking the exponential communication distribution found in actual automotive systems.
Key Results
Comparative testing against CAT, SPA (a heuristic algorithm), and Integer Linear Programming (ILP) variants demonstrated clear advantages for TDCA:
- Bus Traffic Reduction: For the AIS dataset, TDCA maintained a bus traffic range of 700–1100, significantly lower than CAT (1200–1700) and ILP-based methods.
- ECU Equalization: TDCA achieved the lowest standard deviation in ECU utilization, distributing the processing load more evenly than its competitors.
- Stability and Speed: TDCA consistently delivered deployment plans faster than the more complex TDSCA and SPA algorithms, showing high stability as the number of SWCs and ECUs increased.
Conclusion
TDCA provides a robust solution for the deployment of software components in AUTOSAR architectures. By anchoring deployment on high-degree components, the algorithm effectively minimizes inter-ECU communication and balances hardware load. Future research will focus on incorporating strict end-to-end latency deadlines and addressing heterogeneous hardware environments in advanced vehicle architectures.
Summary updated