DeepWeave: Accelerating Job Completion Time with Deep Reinforcement Learning-based Coflow Scheduling

Authors: Penghao Sun, Zehua Guo, Junchao Wang, Junfei Li, Julong Lan, Yuxiang Hu

IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed scheme is evaluated with a simulator using real-life traces. Simulation results show that Deep Weave completes jobs at least 1.7 faster than the state-of-the-art solutions.
Researcher Affiliation Collaboration Penghao Sun1 , Zehua Guo2 , Junchao Wang1 , Junfei Li1 , Julong Lan1 and Yuxiang Hu1 1National Digital Switching System Engineering & Technological R&D Center 2Beijing Institute of Technology sphshine@126.com, guolizihao@hotmail.com, wangjunchao11@126.com, lijunfei90@qq.com, {ndscljl, huyx}@126.com
Pseudocode Yes Algorithm 1 Training process of Deep Weave
Open Source Code No The paper does not provide concrete access to source code for the methodology described in this paper.
Open Datasets Yes In this paper, we use TPC-DS [tpc, 2019] to evaluate the performance of different schemes in job DAGs.
Dataset Splits No The paper mentions training the DRL agent, but it does not provide specific details on dataset splits (e.g., percentages or counts for training, validation, and testing) for its own model.
Hardware Specification Yes The simulation runs on a desktop computer equipped with an Intel i7700 CPU, GTX 1080Ti Graphics card, and 32G DDR4 RAM.
Software Dependencies Yes The DRL algorithm is implemented on Tensor Flow based on Python 3.6.
Experiment Setup Yes In our training, the learning rate α is set to 1 10 3, and the gradient descent is used for neural network parameter update with Adam optimizer [Kingma and Ba, 2014].