SpaceWire Meets Big Data - Realtime Data Mining

Abstract

Abstract —The whole is more than the sum of its parts. This saying particularly fits in the context of big data and data mining. A single data set is only valid and relevant at exactly one point in time. Dependencies between the individual elements of the data set as well as the behaviour over time can only be investigated if large amounts of data are available. A good example is traffic monitoring based on GPS data, WiFi, and/or cellular triangulation of modern mobile phones (see [1], [2]). Here, the data sets are collected from multiple sources to estimate a precise traffic situation. Data completeness is vital to achieving sufficient reflection of the actual state of a system, since information can be lost through downsampling. This is a difficult requirement to fulfill especially for highly dynamic systems like robots. Data recording over the same communication infrastructure from multiple sources must meet strict real time requirements of the control application. Main control loops at joint level of ≥ 3 kHz and up to 100 kHz at motor control are quite common (e.g. [3]). Due to this, a simple duplication of the cyclic data sets at CPU-level for simultaneously controlling and recording is not possible without decimation of the recording sample rate. Therefore recording has to be performed at a different point of the communication infrastructure. This approach provides a duplication of the data at SpaceWire packet level by using seperate SpaceWire regions for control and recording without violating the real time requirements of the control application.