Browse Prior Art Database

Method for Performing Parallel Time Series Transforms and Manipulations on Distributed Time Series Data

IP.com Disclosure Number: IPCOM000249590D
Publication Date: 2017-Mar-07
Document File: 4 page(s) / 76K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method is disclosed for performing parallel time series transforms and manipulations on distributed time series data.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

1

Method for Performing Parallel Time Series Transforms and Manipulations on Distributed Time Series Data

Time series analysis is a popular means for extracting useful knowledge from raw sensor data. With time series analysis, one must manipulate/transform the time series in order to extract the useful data. Currently, executing this transform would require access to the entire time series, since many of these transforms act as sliding windows over the time series. It is advantageous to distribute a time series amongst different nodes, and run a transform in parallel on this time series. There needs a mechanism that run transforms in parallel on time series thereby taking advantage of optimizations that come with parallelism.

Disclosed is a method and system for performing parallel time series transforms and manipulations on distributed time series data. The method transforms a single time series efficiently, and splits the time series over multiple nodes (partitions) that are stored sequentially, with no loss of precision. Subsequently, each partition of the multiple sequential partitions provides first (w-1) elements, where w is sliding window size for the transform to be executed. Thereafter, the method treats each set of elements in the partition including the w-1 elements from a next partition as a single time series and transforms each time series, starting at a correct step interval and preserves ordering of partitions.

In accordance with the method, a single time series is initiated in one node, where each value in the time series is denoted in the form of (<timestamp>, <value>), as illustrated in Figure 1.

Figure 1

2

Figure...