CSR-EHCS(EHS), SM: Development of SYMBIOTE, A Reconfigurable Logic Assisted Data Stream Management System for Multimedia Sensor Networks

Indiana University, Bloomington IN

Investigators

Abstract

This research is developing a novel class of Data Stream Management Systems (DSMSs) called Reconfigurable Logic Assisted DSMS (RLADSMS). The project is developing a prototype called SYMBIOTE to demonstrate real-time processing of high bandwidth multimedia data streams in a computer vision based traffic monitoring system. Using hardware-assisted designs and decomposition strategies to achieve its stream storage model, this project introduces new query operators such as the split-choose and merge operators, and a macro-operator called split-merge. It is building a library of such DSMS operators, realizable in reconfigurable logic hardware, that are reusable, parameterizable, and exhibit reduced run-time and reduced spatial complexity Example hardware-accelerated operators include Filter, Map, Union, Bsort, Aggregate and Join. The project also investigates hardware implementations of strategies such as approximate query answering for sliding windows; landmark windows; and sketches, random sampling, and wavelets. This project extends the Borealis query language of ?boxes and arrows? with the additional intent to address deployment characteristics in SYMBIOTE. The objective is to perform static query optimization using a heuristic best-effort static query planning and diagram placement algorithm that together can minimize the dynamic configuration effort of RLADSMS. This optimization strategy uses a novel cost model for RLADSMS based on the spatial complexity of hardware DSMS operators, run-time complexity of hardware/software operators, and many SYMBIOTE sensor node properties. Sensor node properties can include camera characteristics, frame-rate, number of look-up tables (LUTs)/logic cells, and embedded block RAMs. The optimization can also consider communication properties such as interconnect latencies and bandwidth. The research outcomes of this project are expected to be widely applicable to multimedia sensor network based DSMSs, such as border patrol and building security applications, and possibly to the financial data processing domain. Moreover, this research spurs the academic exploration of novel ideas and creation of new courses in the area of hardware/software codesign of DSMSs.