Supporting Query by Content on ECG Data with User Defined Functions

Abstract

Similarity search, or query by content, is an important operation for time series databases. While the research community has been quite active in this area, we are yet to see full support for these operations and data from commercial Database Management Systems (DBMS). In this work we explore how efficiently can similarity search algorithms be implemented inside a DBMS with User Defined Functions (UDFs). We concentrate our work on querying Electrocardiograms (ECG), a biosignal and a particular type of time series data. Given the lack of support for time series as an standard data type, we identify two alternatives for managing a database of ECG signals using a DBMS, namely using either references to flat files on the operating system, or using Binary Large Object (BLOB) attributes. Our experiments show a significant overhead in the total elapsed time while doing similarity search on signals stored as BLOB. On the other hand, querying signals stored as files using UDFs is as competitive as using an ad-hoc implementation of the query by content algorithm running as a stand-alone application on the operating system.