Print Email Facebook Twitter Performance Analysis of Middleware for Component Based Robot Control Software Title Performance Analysis of Middleware for Component Based Robot Control Software Author Van der Hoorn, G. Contributor Gross, H.G. (mentor) Wisse, M. (mentor) Faculty Electrical Engineering, Mathematics and Computer Science Department Software Engineering Programme Computer Science Date 2012-08-29 Abstract The design and implementation of robot control applications is complex: the enormous diversity in hardware, people and software leads to issues with productivity, maintainability and reusability. To cope with these challenges, robotics is adopting software engineering techniques, and in particular Component Based Software Engineering (CBSE). With the main rationale for this transition being CBSE's many advantages, the impact on the performance and efficiency of control applications has not received a similar amount of attention. In light of this, this study investigates the effects of componentisation on the performance of robot control applications, with particular attention to the influence on latency, jitter and throughput. The evaluation consists of two parts: first, synthetic benchmarks are used to assess the capabilities of two popular robotics middlewares -- the Robot Operating System (ROS) and the Open RObot COntrol System (OROCOS), selected for their support for the dataflow, publish-subscribe and remote procedure call interaction styles. The second part consists of a case-study, in which a monolithic AIS message decoder is componentised and implemented on both middlewares. Then, six variants of the resulting component system -- one for each of the interaction styles, plus two special configurations -- are benchmarked, and the effects on throughput, latency and jitter are analysed. Synthetic benchmark results indicate a strong correlation between message size and interaction latency, due to significant serialisation overhead in both middlewares. Case-study results show a significant negative influence on the performance metrics. Maximum system throughput is reduced by 99% in the worst-case and 94% in the best-case. Compared to the monolithic system, end-to-end latency increases between 7.5 and 1700 times at maximum throughput. Jitter increases from 0.1us to 24us for the OROCOS dataflow system, and to 980us for the ROS publish-subscribe system. Based on these results, componentisation of the system in the case-study cannot be recommended in contexts where performance is important. Subject middlewareperformanceroboticscomponents To reference this document use: http://resolver.tudelft.nl/uuid:4e36121c-da15-46ab-953e-d6a87d1d4271 Embargo date 2012-08-31 Part of collection Student theses Document type master thesis Rights (c) 2012 Van der Hoorn, G. Files PDF 20120821_msc-thesis_gvdhoorn.pdf 1.7 MB Close viewer /islandora/object/uuid:4e36121c-da15-46ab-953e-d6a87d1d4271/datastream/OBJ/view