Sensors are pervasively deployed on mobile devices with the development of Internet of Things technology. Value-added services are innovated and developed by analyzing data streams from massive number of mobile sensors in online mode. Due to dynamic working condition of mobile sensors and the high data rate, back end analytic services confront incoming streams with large rate fluctuation and out-of-order time series. This puts forward special challenges in service implementation for commercial applications, where good reliability/scalability performance is a must. In this paper, a data ingestion and scheduling framework is proposed to enable large-scale tempo-spatial streams analysis in a reliable and cost-effective way. A case study on a real world application adopting this framework is introduced and its pilot result is presented.