リアルタイム実行環境におけるオープンソース出版/購読型通信ミドルウェアの性能評価

抄録

This paper evaluates the performance of four open-source publication/subscription communication middlewares, Fast DDS, Cyclone DDS, iceoryx, and eCAL, on shared memory inter-process communication in a real-time execution environment. I set up two hypothetical robot systems and measured the communication latency and publication processing duration of these communication middlewares in the real-time execution environment. Experimental results for the first robot system (with a payload of 1kB and a publication interval of 1ms) showed that the communication middleware with the smallest communication latency is iceoryx. Apart from the quantitative evaluation of communication latency, qualitative aspects such as single point of failure, communication with other computers, and availability of GUI monitoring tools are also considered. The results show that eCAL is a well-balanced option with the second lowest maximum communication delay after iceoryx. In the second robot system, which is a mobile robot navigation robot system, iceoryx and eCAL, which performed well in the motion control robot system experiments, are evaluated in more detail with payloads up to 10MB. iceoryx and eCAL support the Zero-Copy function, which reads messages directly from shared memory when subscribing to a message. This function eliminates the need for copying by communication. However, the measured communication latency and publication processing duration show that iceoryx requires only one copy when publishing a message, while eCAL requires two copies. Accordingly, eCAL requires more communication latency and publication processing duration than iceoryx. The performance of eCAL is also measured for Byte-string and Protocol Buffers message formats, and the results show that serialization significantly impacts performance. For relatively large messages with payloads of 10MB, iceoryx performed 111 times better than eCAL regarding maximum communication delay and three times better in maximum publish processing duration. These results indicate that iceoryx is the most suitable for inter-process communication with large payloads.