使用可插拔和可定制的智能运行时提升工作负载的QoS | KC23

Enhance Workload QoS with Pluggable and Customizable Smarter Runtimes - Rougang Han & Kang Zhang

使用可插拔和可定制的智能运行时提升工作负载的QoS | Enhance Workload QoS with Pluggable and Customizable Smarter Runtimes - Rougang Han, Alibaba & Kang Zhang, Intel

随着云基础业务类型和硬件资源的日益丰富，数据中心的资源利用率得到了显著提高，但也带来了资源争用的风险。在提高节点资源利用率的同时，确保应用程序的QoS（服务质量）而不妥协，并避免嘈杂的邻居问题是一个关键挑战。本主题介绍并演示了如何通过在Kubernetes上集成最新的CRI运行时（例如containerd、cri-o）中的NRI（节点资源接口），在Koordinator中确保工作负载的QoS。与传统方法（如独立和运行时代理拦截CRI请求）相比，NRI的插件化设计避免了对Kubelet的侵入性修改，极大增强了Koordinator部署的灵活性，实时处理Pod生命周期事件，并为云原生系统提供了一种优雅且标准化的资源管理解决方案。

With increasingly abundant cloud-based business types and hardware resources, the resource utilization of data centers has dramatically improved but also brings the risk of resource contention. While enhancing the utilization of node resources, one of the critical challenges is to ensure application QoS (Quality of Service) without compromise and avoid the Noisy Neighbor problem. This topic introduces and demonstrates how to ensure workload QoS in the Koordinator, an open-source scheduling system on Kubernetes, by incorporating the latest NRI (Node Resource Interface) in CRI runtime (e.g. containerd, cri-o). Compared to traditional approaches, such as standalone and runtime proxy to intercept CRI requests, the plugin-based design of NRI avoids invasive modifications to Kubelet, greatly enhancing the flexibility of Koordinator deployment, real-time processing of pod lifecycle events and providing an elegant and standardized resource management solution for cloud-native systems.

CNCF概况（幻灯片）

扫描二维码联系我们！

CNCF (Cloud Native Computing Foundation)成立于2015年12月，隶属于Linux  Foundation，是非营利性组织。 

CNCF（云原生计算基金会）致力于培育和维护一个厂商中立的开源生态系统，来推广云原生技术。我们通过将最前沿的模式民主化，让这些创新为大众所用。请关注CNCF微信公众号。