arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2410.15853cs.PF
收藏

ADS性能再探

ADS Performance Revisited

Alexander Weber, Holger Eichelberger, Jobst Hildebrand

AI总结:

本文重新分析Beckhoff ADS协议在工业4.0实时边缘通信中的性能瓶颈,证明Java-JNI和C++直接集成均无法超越Java-JNA,且主要耗时在ADS本身,并提出ADS库的替代使用方式可提升性能。

摘要:

实时测量对于制造过程的深度控制非常重要,对于现代AI方法而言,这需要与高级语言集成。在我们上一篇SSP论文中,我们研究了一种基于Python和一种基于Java-JNA的方法,将Beckhoff ADS协议用于实时边缘通信并集成到工业4.0平台中的性能。在那里,我们已经表明,虽然Java优于Python,但两种解决方案都未达到根据任务而定的1-20kHz的期望目标。然而,我们仍然缺乏对这一结果的解释以及对替代方案的分析。针对第一个主题,我们在本文中表明:1)在该场景中将Java-JNA替换为Java-JNI并不能进一步提高性能;2)一个以更直接集成方式实现相同行为的C++程序表现并不更好;3)性能分析显示,大部分执行时间都花费在ADS上。针对第二个主题,我们表明对ADS库的替代使用方式可以实现更好的性能。

英文摘要:

Real-time measurements are important for in-depth control of manufacturing processes, which, for modern AI methods, need integration with high-level languages. In our last SSP paper we investigated the performance of a Python and a Java-JNA based approach to integrate the Beckhoff ADS protocol for real-time edge communication into an Industry 4.0 platform. There, we have shown that while Java outperforms Python, both solutions do not meet the desired goal of 1-20kHz depending on the task. However, we are are still lacking an explanation for this result as well as an analysis of alternatives. For the first topic, we show in this paper that 1) exchanging Java-JNA with Java-JNI in this setting does not further improve the performance 2) a C++ program realizing the same behavior in a more direct integration does not perform better and 3) profiling shows that the majority of the execution is spend in ADS. For the second topic, we show that alternative uses of the ADS library allow for better performance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑