CARES: code and raw results for "Completion-Aware, Rejection-Enabled Deep Reinforcement Learning for Deadline-Constrained Task Allocation in Volatile Fog Computing"
Authors/Creators
- 1. Krishna Institute of Engineering and Technology (KIET), Ghaziabad, India
- 2. Manipal University Jaipur, Jaipur, India
Description
Code, trained agents and raw results for the paper "CARES: Completion-Aware, Rejection-Enabled Deep Reinforcement Learning for Deadline-Constrained Task Allocation in Volatile Fog Computing" (submitted to Cluster Computing).
The archive contains a deterministic discrete-event fog simulator (validated against the M/G/1 Pollaczek-Khinchine formula), the CARES scheduler (double DQN with a shared, permutation-invariant candidate scorer, an explicit admission action and an outcome-based reward), 15 baseline schedulers, the training, evaluation, statistics and plotting scripts, the raw per-seed metrics of every evaluation run, the 70 trained CARES agents, and a processed version of the Azure Functions Invocation Trace 2021 used for the trace-driven evaluation.
Every table and figure of the paper can be rebuilt from this archive without re-running the experiments (see README.md). Code: MIT licence. Results: CC BY 4.0. The processed trace is derived from the Azure Functions Invocation Trace 2021 (Microsoft, CC BY 4.0).