Centaurus v0.5 Release Announcement
Mengni Zhang
Hi all,
With continuous support from the Centaurus open source community, we are very pleased to announce that we just cut the v0.5 release of Centaurus. In Centaurus v0.5, we have included release features from all four of Centaurus SIGs: Scalability SIG (Arktos) , Networking SIG (Mizar), Edge SIG (Global Scheduler) and AI SIG (Alnair). This release also marks the first official release for Global Scheduler and Alnair sub-project.
Release overview:
- Scalability: Arktos v0.8 - Arktos scale-out architecture can now achieve 50K-node cluster scale along with inherent secure communication support.
- Networking: Mizar v0.8 – Mizar now supports Label-based Network Policy for K8S and Bandwidth & QoS for pod network traffic.
- Edge and Global Scheduler: Global Scheduler v0.1 - The first release introduces the design and implementation of the components of the global scheduling platform with performance improvements and validation. Edge SIG also introduces the design of Centaurus Edge backed by POC work.
- AI: Alnair v0.1 - The first release introduces two components: Profiler and Elastic Training Framework.
Detail features of the release as follows:
Arktos v0.8 Arktos scale-out architecture can now achieve 50K-node cluster scale along with inherent secure communication support.
In Arktos v0.8, we continue to improve scalability with the new scale-out architecture we introduced in Arktos v0.7. Also, Arktos can now support multiple resource managers where tenant can utilize resources across all the resource managers. Together, Arktos successfully passed the 50K-node density test with maximum 6s start up latency. We will continue to improve the scale and performance of Arktos in future releases.
Please refer to https://github.com/CentaurusInfra/arktos/releases/tag/v0.8 for a full list of features in this release.
Mizar v0.8 Mizar now support a native-label based enforcement mechanism which greatly reduces update cost when label selection changes.
In the previous release Mizar provided full support of Kubernetes network policy. Under the hood, Mizar translates Kubernetes label selections to IP groups to perform policy enforcement. In Mizar v0.8 release, we now support a native label-based enforcement mechanism, which avoids label-IP translation and greatly reduces update cost when label selection changes. In addition, this release also includes network QoS control for different priority pods, major stabilization fixes and single-yaml Mizar deployment for a more user-friendly experience.
Please refer to https://github.com/CentaurusInfra/mizar/releases/tag/v0.8 for a full list of features in this release.
Global Scheduler v0.1 The first release introduces the design and implementation of the components of the global scheduling platform with performance improvements and validation.
Global Scheduler is an open source large scale cloud resource orchestration and scheduling platform. It aims to address key scheduling challenges of compute units across a large number of DC clouds and large clouds. In this release, we complete the design and code development of all the components of the global scheduling platform. We also optimize code design/implementation to support 1000 clusters testing scenarios, 100 pods/s QPS, and 100 allocations/s.
We have incorporated global scheduler work as part of the Edge SIG. Please refer to https://github.com/CentaurusInfra/global-resource-scheduler/releases/tag/v0.1 for a full list of features in this release.
Additionally, Edge SIG also brings the design of the Centaurus Edge backed by Proof of Concept implementation. Please visit the design document here https://github.com/pdgetrf/ArktosEdge/blob/main/design/530_design.md.
- Proof of Concepts: https://github.com/CentaurusInfra/fornax, and key items implemented as POC:
- Cascading Edge Cluster
- Workload Assignment
- Workload and edge cluster status reporting
- Support multiple K8s flavors as edge clusters
Alnair v0.1 The first release introduces two components: Profiler and Elastic Training Framework
Alnair aims to build an intelligent platform to improve AI workloads efficiency. AI workloads are very critical for cloud and edge computing. With Centaurus AI SIG, we will focus on the resource management aspects, to analyze and schedule AI workloads on existing/new systems with intelligent methods.
Please refer to https://github.com/CentaurusInfra/alnair/releases/tag/v0.1 for a full list of features in this release.
Future Outlook
With Centaurus v0.5 release, Centaurus can now achieve higher scalability with better stabilization. This release also brings in a much more friendly user experience and better community contributor experience. The addition of research work from Edge and AI SIGs also highlights more possibilities as we work towards addressing cloud computing challenges in the age of 5G, AI and Edge. A big thanks to our partners and community for their continuous support.
For the upcoming releases, the Centaurus community will continue to improve the scale and performance of Centaurus Platform. The Edge SIG will begin implementation of the Centaurus Edge design. The AI SIG will introduce new architecture on how to orchestrate heterogeneous resources and new services model to facilitate AI workloads. We are excited about the works that are in the plan. We welcome anyone who's interested in tackling cloud challenges to join the community. For more details regarding Centaurus community, please join us here: https://github.com/CentaurusInfra
Best regards,
Mengni Zhang on behalf of the Centaurus Team
Futurewei Technologies, Inc.
This e-mail and any attachments may contain confidential information from Futurewei, which are intended only for the person or entity whose email address appears above. Any use of the information attached or contained herein in any way (including, but not limited to, total or partial disclosure, reproduction, or any unapproved dissemination) by persons other than the intended recipient(s) is prohibited. If you have received this e-mail in error, please notify the sender by phone or response email immediately and delete this original message.