Hadoop. TS: Large-Scale Time-Series Processing

Abstract

The paper describes a computational framework for time-series analysis. It allows rapid prototyping of new algorithms, since all components are re-usable. Generic data structures repre-sent different types of time series, e. g. event and inter-event time series, and define reliable interfaces to existing big data. Standalone applications, highly scalable MapReduce pro-grams, and User Defined Functions for Hadoop-based anal-ysis frameworks are the major modes of operation. Effi-cient implementations of univariate and bivariate analysis al-gorithms are provided for, e. g., long-term correlation, cross-correlation and event synchronization analysis on large data sets.