Extensible Data Analytics System

Abdul Rehman

Advisor: Prateek Sharma

Indiana University, Bloomington

Table of Contents

• Overview
• Architecture
• Presentation
• Performance Results
• Repository Layout
• Getting Started

Overview

Edge-IoT-Analytics-Box is a data analytics system that captures and analyzes data in real time. The system is designed to be deployed on edge devices like Jetson Orin.

Architecture

Presentation

Performance Results

For 54 sensors being predicted using arima model and RNN model

• Network bandwidth consumption on Jetson ~800 Kbytes per sec
• CPU utilization on Jetson ~25 %
• GPU utilization ~40 %
• InfluxDB storage increases by ~7 Kbytes per second
  • it would take 866 days to fill up 500 Gb of space which is ~2.3 years

Please see report for more details.

Repository Layout

Each module has it's own directory.

./tserve (Torch Serve): Contains code to deploy torch serve on Jetson Orin along with scripts for building and delploying models.

./analytics (Framework): Framework to setup a data analytics pipeline using models deployed on torch serve.

./streaming_service (Data Fetching): Contains code to setup data streaming from sensors to InfluxDB, including scripts to setup InfluxDB.

Getting Started

• setup InfluxDB on a server using the instructions in streaming_service/README.md
• once data is streaming to the DB
  • setup torch serve on a Jetson Orin using the instructions in tserve/README.md
  • setup an analytics pipeline using the instructions in analytics/README.md