English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Coflow is a recently proposed networking abstraction to help improve thecommunication performance of data-parallel computing jobs. In multi-stage jobs,each job consists of multiple coflows and is represented by a Directed AcyclicGraph (DAG). Efficiently scheduling coflows is critical to improve thedata-parallel computing performance in data centers. Compared with hand-tunedscheduling heuristics, existing work DeepWeave [1] utilizes ReinforcementLearning (RL) framework to generate highly-efficient coflow scheduling policiesautomatically. It employs a graph neural network (GNN) to encode the jobinformation in a set of embedding vectors, and feeds a flat embedding vectorcontaining the whole job information to the policy network. However, thismethod has poor scalability as it is unable to cope with jobs represented byDAGs of arbitrary sizes and shapes, which requires a large policy network forprocessing a high-dimensional embedding vector that is difficult to train. Inthis paper, we first utilize a directed acyclic graph neural network (DAGNN) toprocess the input and propose a novel Pipelined-DAGNN, which can effectivelyspeed up the feature extraction process of the DAGNN. Next, we feed theembedding sequence composed of schedulable coflows instead of a flat embeddingof all coflows to the policy network, and output a priority sequence, whichmakes the size of the policy network depend on only the dimension of featuresinstead of the product of dimension and number of nodes in the job'sDAG.Furthermore, to improve the accuracy of the priority scheduling policy, weincorporate the Self-Attention Mechanism into a deep RL model to capture theinteraction between different parts of the embedding sequence to make theoutput priority scores relevant. Based on this model, we then develop a coflowscheduling algorithm for online multi-stage jobs.

A Scalable Deep Reinforcement Learning Model for Online Scheduling Coflows of Multi-Stage Jobs for High Performance Computing