Skip to content

An implementation of Graphene, a research-oriented job scheduler from Microsoft

Notifications You must be signed in to change notification settings

stxue1/graphene

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

An implementation of Graphene: Packing and Dependency-aware Scheduling for Data-Parallel Clusters

Abstract

The modern landscape of cluster computing often deals with increasingly complex workloads characterized by DAG-structured jobs. One recent trace that highlights these complex characteristics is the 2018 Alibaba cluster trace. These modern workloads warrant a re-evaluation of existing algorithms to determine their efficacy in real-world scenarios. Graphene is a near-optimal algorithm for scheduling Directed Acyclical Graph (DAG) workflows. In its original evaluation with a synthetic workload, it achieved up to a 50% reduction in job completion time compared to a critical path scheduler. However, Graphene bases its claims on synthetically generated trace data. Since real world trace data is now available, such as the Alibaba cluster traces, this allows researchers to answer more complicated questions in cluster scheduling. Therefore, we analyzed concerns like queueing delay in the Alibaba cluster, then re-evaluated Graphene on the Alibaba trace to demonstrate its efficacy on real-world trace data.

About

An implementation of Graphene, a research-oriented job scheduler from Microsoft

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published