Networked Edge Computing: Models, Scheduling Algorithms, and Hands-on Evaluation

Presenters

Presenter	Institution
Bhaskar Krishnamachari	University of Southern California
Jared Coleman	Loyola Marymount University

Bhaskar Krishnamachari (bkrishna@usc.edu)
Professor, University of Southern California

Jared Coleman (jared.coleman@lmu.edu; primary contact)
Assistant Professor, Loyola Marymount University

Format and Duration

Half-day (~3 hours), lecture + hands-on workshop hybrid.

Summary

As edge computing and IoT systems become increasingly prevalent, scheduling complex, graph-structured applications across heterogeneous networked compute nodes has emerged as a central challenge at the intersection of networking and distributed systems. This half-day tutorial provides both a conceptual foundation and practical tools for understanding and evaluating task scheduling in edge compute networks. The first half covers models for task graph scheduling on heterogeneous networks, classical and modern heuristics (e.g., HEFT, CPOP, ETF), and real-world deployment considerations drawn from dispersed computing architectures such as Jupiter. The second half is a hands-on workshop using SAGA, a modular, open-source Python framework supporting over 20 scheduling algorithms, where attendees will implement schedulers, benchmark them on realistic workloads, use PISA (a simulated annealing-based adversarial analysis tool) to stress-test algorithms and expose failure cases, and use NCSIM (a wireless networked computing simulator) to evaluate their performance on realistic wireless edge networks. Attendees will leave with both a conceptual understanding of edge scheduling challenges and practical experience with reproducible tools for algorithm evaluation.

Tutorial Timetable

Part 1	Models and Algorithms (90 min)
20 min	Introduction and Motivation: why scheduling matters for edge networks; real-world applications; NP-hardness; how network characteristics affect scheduling.
20 min	Task Graph Scheduling: formal problem definition, key parameters (CCR, heterogeneity), etc.
10 min	Interactive Exercise: pencil-and-paper DAG scheduling on a heterogeneous network.
20 min	Classical and Modern Heuristics: HEFT, CPOP, ETF, parametric frameworks, GCN-based learned schedulers, comparison with Kubernetes/KubeEdge and Ray.
20 min	Deployment and Practical Considerations: dispersed computing (Jupiter), online scheduling under uncertainty, lessons from tactical edge environments.
Part 2	Hands-on with SAGA (90 min)
15 min	Getting Started with SAGA via pre-configured GitHub Codespace.
25 min	Implementing and Benchmarking Schedulers using SAGA's modular interface.
15 min	Adversarial Analysis with PISA: finding instances where heuristics fail.
20 min	Scheduling on Wireless Edge Networks with NCSIM: evaluating idealized schedules under realistic wireless conditions.
15 min	Wrap-up and Discussion: open challenges, SAGA as a benchmarking platform, Q&A.

Expected Audience and Prerequisites

Designed for researchers and practitioners interested in edge computing, IoT scheduling, and distributed systems. Attendees should have a basic understanding of networking concepts (latency, bandwidth, distributed systems) and Python programming experience. Experience with scheduling is helpful but not required. We will cover the necessary background on scheduling models and algorithms.

Laptop Requirements

Any reasonably modern laptop/OS. Attendees are encouraged to use a free GitHub Codespace for the hands-on session (coding in under a minute, no installation needed). Local installation is also supported on Linux, macOS, and Windows (via WSL/Docker).

Artifact Availability

All tutorial materials (slides, Jupyter notebooks, and code) will be publicly available via the SAGA GitHub repository: https://github.com/ANRGUSC/saga.

Biographies

Bhaskar Krishnamachari is a Professor of Electrical and Computer Engineering at the University of Southern California and directs the USC Center for Cyber-Physical Systems and the Internet of Things. An IEEE Fellow, he has co-authored over 400 publications with 37,000+ citations, spanning IoT, edge computing, wireless networks, and distributed systems. His group developed the Jupiter platform for dispersed computing, the SAGA framework for scheduling, and the NCSIM simulator.

Jared Coleman is Assistant Professor at Loyola Marymount University and the lead developer of the SAGA framework. His research focuses on task scheduling algorithms for distributed and edge systems, with publications at IPDPS, JSSPP, MILCOM, and SPIE. He has applied scheduling research in NATO tactical edge environments and HPC contexts. His work in scheduling is currently supported in part by the National Science Foundation under Award No. 2451267.