Volcano

Cloud native batch scheduling system for compute-intensive workloads

Network Topology Aware Scheduling

Supports network topology aware scheduling, significantly reducing application communication overhead between nodes, and greatly enhancing model training efficiency in AI distributed training scenarios

Online and Offline Workloads Colocation

Supports online and offline workloads colocation, enhancing resource utilization while ensuring online business QoS through unified scheduling, dynamic resource overcommitment, CPU burst capabilities, and resource isolation

Multi Cluster Job Scheduling

Supports cross cluster job scheduling for larger scale resource pool management and load balancing

Star

Learn more about volcano multi-cluster scheduling

Hierarchical Queue Management

Supports multi-level queue management for fine-grained resource quota control

Descheduling

Supports dynamic descheduling to optimize cluster load distribution and improve system stability

Star

Learn more about volcano descheduling

Multiple Scheduling Policies

Supports various scheduling strategies including Gang, Fair-Share, Binpack, DeviceShare, Capacity, Proportion, NUMA aware, and Task Topology, optimizing resource utilization

Heterogeneous Device Support

Supports scheduling for various heterogeneous devices like GPU and NPU, unleashing hardware computing power

High Performance Unified Scheduling

Supports native Kubernetes workload scheduling while providing complete support for frameworks like Ray, PyTorch, TensorFlow, Spark, and Flink through VolcanoJob, achieving unified job scheduling with excellent performance

Powerful Monitoring

Provides rich logging, monitoring metrics, and dashboards

Star

Learn more about volcano dashboard

About Volcano

Volcano

A cloud native system for high-performance workloads

Volcano is CNCF’s first cloud native batch computing project, focusing on high performance computing scenarios such as AI, big data, and genomics analysis. Its core capabilities include:

• Unified Scheduling: Supports integrated job scheduling for both Kubernetes native workloads and mainstream computing frameworks (such as TensorFlow, Spark, PyTorch, Ray, Flink, etc.).

• Queue Management: Provides multi-level queue management capabilities, enabling fine-grained resource quota control and task priority scheduling.

• Heterogeneous Device Support: Efficiently schedules heterogeneous devices like GPU and NPU, fully unleashing hardware computing potential.

• Network Topology Aware Scheduling: Greatly enhancing model training efficiency in AI distributed training scenarios.

• Multi-cluster Scheduling: Supports cross cluster job scheduling, improving resource pool management capabilities and achieving large scale load balancing.

• Online and Offline Workloads Colocation: Enables online and offline workloads colocation, improving cluster resource utilization through intelligent scheduling strategies.

• Load Aware Descheduling: Optimizing cluster load distribution and enhancing system stability.

As the industry’s first cloud native batch computing engine, Volcano has been widely applied in high-performance computing scenarios such as artificial intelligence, big data, and genome sequencing, providing powerful support for enterprises to build elastic, efficient, and intelligent computing platforms.

Why Volcano

A powerful batch scheduler that allows you to run multi-architecture, computing-intensive jobs as Kubernetes workloads

Spark

Apache Spark™ is a unified analytics engine for large-scale data processing.

Flink

Apache Flink is a framework and distributed processing engine for stateful computations over unbounded and bounded data streams.

TensorFlow

An end-to-end open source machine learning platform.

PyTorch

An open source machine learning framework that accelerates the path from research prototyping to production deployment.

argo

Argo Workflows is an open source container-native workflow engine for orchestrating parallel jobs on Kubernetes. Argo Workflows is implemented as a Kubernetes CRD.

MindSpore

The all-scenario deep learning framework developed by Huawei.

Ray

Ray is a high-performance distributed computing framework that supports machine learning, deep learning, and distributed applications.

Kubeflow

The Kubeflow project is dedicated to making deployments of machine learning (ML) workflows on Kubernetes simple, portable, and scalable.

Open MPI

The Open MPI Project is an open source Message Passing Interface implementation that is developed and maintained by a consortium of academic, research, and industry partners.

Horovod

Horovod is a distributed deep learning training framework for TensorFlow, Keras, PyTorch, and Apache MXNet.

MXNet

A truly open source deep learning framework suited for flexible research prototyping and production.

PaddlePaddle

PaddlePaddle is an open source deep learning platform derived from industrial practice initiated by Baidu.

Recent & Upcoming Talks

Intro: Kubernetes Batch Scheduling

Kevin Wang

Kubernetes started as a general purpose orchestration framework with a focus on serving jobs. But as it gains popularity, users want to run high performance workloads on Kubernetes, such as Spark, TensorFlow, etc. When …

May 22, 2019 2:50 PM — 3:25 PM