SplitPlace is a container orchestration framework for dynamic scheduling and decision making in resource constrained edge environments. SplitPlace decides whether to use semantic or layer wise splits of neural network applications with latency and accuracy critical user requirements on distributed setups with low memory legacy devices.
SplitPlace is based on the COSCO Framework and uses the co-simulation and surrogate optimization primitives of COSCO. To run the framework, install required packages using
python3 install.pyTo run the code with the required scheduler, modify lines 81 and 85 of main.py to one of the several options.
decider = MABDecider()
...
scheduler = GOBIScheduler('energy_latency_'+str(HOSTS))To run the simulator, use the following command
python3 main.pyAccess the wiki for installation instructions and replication of results.
| Items | Contents |
|---|---|
| Paper | https://ieeexplore.ieee.org/document/9780535 |
| Pre-print | https://arxiv.org/pdf/2205.10635.pdf |
| Documentation | https://github.com/imperial-qore/COSCO/wiki |
| Video | (coming soon) |
| Contact | Shreshth Tuli (@shreshthtuli) |
| Funding | Imperial President's scholarship, H2020-825040 (RADON) |
Our work is published in IEEE TMC journal. Cite using the following bibtex entry.
@article{tuli2021splitplace,
author={Tuli, Shreshth and Casale, Giuliano and Jennings, Nicholas R.},
journal={IEEE Transactions on Mobile Computing},
title={{SplitPlace: AI Augmented Splitting and Placement of Large-Scale Neural Networks in Mobile Edge Environments}},
year={2022}
}BSD-3-Clause. Copyright (c) 2020, Shreshth Tuli. All rights reserved.
See License file for more details.