大型集群中由containerd插件驱动的节点级优化 | KC23

Node-Level Optimization in Large Cluster Powered by Containerd Plugin - Tianyi Tang, Alibaba Cloud

大型集群中由containerd插件驱动的节点级优化 | Node-Level Optimization in Large Cluster Powered by Containerd Plugin - Tianyi Tang, Alibaba Cloud

在大规模的Kubernetes集群中，Pod的冷启动时间和噪声邻居效应是不可忽视的。像函数计算这样的场景需要较低的Pod冷启动时间，而镜像分发效率正成为Pod扩展的瓶颈。目前，Pod的IO利用率无法像CPU和内存那样进行限制，IO噪声邻居效应会使延迟敏感的Pod工作负载面临风险。多租户集群中的IO利用率管理为节点管理带来了新的挑战。在本次演讲中，我们将描述如何利用containerd的插件机制、新兴的延迟加载镜像和P2P加速器来提供相应的解决方案和增强功能，如动态镜像垃圾回收插件和读/写磁盘分离。这些解决方案将促进社区发展，并帮助其他Containerd终端用户获得处理镜像分发效率和IO压力相关障碍的开箱即用体验。

In large-scale Kubernetes clusters, pod cold-start time and noise neighbor effects are non-neglectable. Scenarios like Function Compute require low pod cold-start time and image distribution efficiency is growing as the bottleneck in pod scaling. Currently, the pod’s IO utilization cannot be limited as CPU and memory, IO noise neighbor effects will put workloads like latency-sensitive pods at risk. IO utilization management in multi-tenant clusters introduces novel challenges for node management. In this talk, we will describe how we leveraged on containerd's plugin mechanism, emerging lazy-loading images, and p2p accelerators to provide corresponding solutions and enhancements like dynamic image garbage collection plugin and read/write disk separation. These solutions will foster the community and help other Containerd end-users gain out-of-the-box experiences tackling image-distribution efficiency and IO pressure-related obstacles.

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。