Edge microservice applications are becoming a viable solution for the execution of real-time IoT analytics, due to their rapid response and reduced latency. With Edge Computing, unlike the central Cloud, the amount of available resource is constrained and the computation that can be undertaken is also limited. Microservices are not standalone, they are devised as a set of cooperating tasks that are fed data over the network through specific APIs. The cost of processing these feeds of data in real-time, especially for massive IoT configurations, is however generally overlooked. In this work we evaluate the cost of dealing with thousands of sensors sending data to the edge with the commonly used encoding of JSON over REST interfaces, and compare this to other mechanisms that use binary encodings as well as streaming interfaces. The choice has a big impact on the microservice implementation, as a wrong selection can lead to excessive resource consumption, because using a less efficient encoding and transport mechanism results in much higher resource requirements, even to do an identical job.

由于其快速响应和减少延迟，边缘微服务应用程序正在成为执行实时物联网分析的可行解决方案。与中央云不同，边缘计算的可用资源量受到限制，可以进行的计算也受到限制。微服务不是独立的，它们被设计为一组协作任务，这些任务通过特定的 API 通过网络提供数据。然而，实时处理这些数据源的成本，尤其是对于大规模物联网配置，通常被忽视。在这项工作中，我们评估了处理通过 REST 接口使用常用的 JSON 编码向边缘发送数据的数千个传感器的成本，并将其与使用二进制编码和流接口的其他机制进行比较。