Cucumber

A configurable admission control for resource-constrained compute nodes with on-site renewable energy sources.

Cucumber accepts delay-tolerant workloads in a way that increases renewable excess power utilization through probabilistic multistep-ahead forecasts of computational load, energy consumption, and energy production.

This repository contains datasets, simulation code, and analysis notebooks used in the evaluation of the paper "Cucumber: Renewable-Aware Admission Control for Delay-Tolerant Cloud and Edge Workloads".

Data

The data directory contains all input data for the simulation.

Capacity Forecasts

data/u_<scenario>_<solar_site> contains the capacity forecasts for a <scenario> x <solar_site> combination. Each line represents a forecast result and is indexed by a multi-index where level 1 is the datetime of the forecast request and level 2 the datetime of the forecast result.

For each forecast result the following information is available:

• u_free: Actual free capacity
• u_free_pred: Forecasted free capacity
• u_reep: Actual capacity which can be powered by renewable excess energy
• u_reep_pred_expected: Forecasted capacity which can be powered by renewable excess energy (expected case)
• u_reep_pred_conservative: Forecasted capacity which can be powered by renewable excess energy (conservative case)
• u_reep_pred_optimistic: Forecasted capacity which can be powered by renewable excess energy (optimistic case)

Workload Requests

data/requests_<scenario> contains the requests for delay-tolerant workloads that Cucumber can accept/reject during the simulation. Each line contains the arrival time, deadline, and size of the workload request.

Simulation and Analysis

The evaluation directory contains a SimPy simulation used for evaluating Cucumber. To run the simulation, you need Python 3.7 or newer and install the dependencies in requirements.txt, e.g. via:

python3 -m venv venv  			# create venv
. venv/bin/activate   			# activate venv
pip3 install -r requirements.txt	# install dependencies

Now execute the simulation via:

cd evaluation
python main.py

All output files of this step are in directory results:

• results/u_<scenario>_<solar_site>_<policy>: Capacity usage statistics for each experiment (e.g. available spare capacity, green excess capacity, actually used capacity)
• results/jobs_<scenario>_<solar_site>_<policy>: Workload statistics for each experiment (e.g. arrival time, finish time, deadline)

The analysis.ipynb notebook analyzes the results and generates the plots presented in the paper, you can start it via:

jupyter lab

Publication

Cite as:

• Philipp Wiesner, Dominik Scheinert, Thorsten Wittkopp, Lauritz Thamsen, and Odej Kao. "Cucumber: Renewable-Aware Admission Control for Delay-Tolerant Cloud and Edge Workloads" In the Proceedings of the 28th International European Conference on Parallel and Distributed Computing (Euro-Par). Springer. 2022. [arXiv preprint]

Bibtex:

@inproceedings{Wiesner_Cucumber_2022,
  author={Wiesner, Philipp and Scheinert, Dominik and Wittkopp, Thorsten and Thamsen, Lauritz and Kao, Odej},
  booktitle={28th International European Conference on Parallel and Distributed Computing (Euro-Par)}, 
  title={Cucumber: Renewable-Aware Admission Control for Delay-Tolerant Cloud and Edge Workloads}, 
  publisher={Springer}
  year={2022}
}