In recent years, the evolving of IoT (Internet of Things) has resulted in the deployment of massive numbers of sensors in various fields, such as the Energy and Utility (E&U) industry. These sensors are continuously producing a huge amount of time series data, which creates a correspondingly huge demand for time series data analysis, such as pattern searching. However, analysis on time series data is currently implemented as custom applications, a strategy that suffers from low efficiency and high maintenance costs. Hence there is a need to provide a service for analysis on time series data that reduces maintenance costs and enhances query efficiency. Existing time series data management services lack the capability to perform pattern searches on the massive amount of data from sensors. This paper presents Time Series analytics as a Service (TSaaaS), a scalable analytic service for time series data in IoT scenarios. We designed pattern searching in TSaaaS that can support efficient and effective searching on truly massive amounts of time series data with very little overhead on the IoT system. To simplify access to the TSaaaS, we created a group of RESTful web interfaces. TSaaaS is implemented as an extension to the Time Series Database service in the IBM cloud platform offering (Codename: BlueMix), which is a new product to accelerate IoT application development. TSaaaS targets a future release of the Time Series Database service. We have conducted proofs of concept (PoC) of TSaaaS with real-world customers from power meter management and bridge monitoring in China. The pilot results and other experiments show that for a selection of pattern cases provided by customers, pattern searches via our service are 10-100 times faster than the existing techniques, while the additional storage cost for the service provider accounts for only about 0.4% of original time series data.

TSAaaS: Time Series Analytics as a Service on IoT