Loading…

线上峰会
12月9-10日
了解更多信息注册参加

Sched 应用程式允许你建立你的日程表,但不能代替你的活动注册。你必须注册 2021年中国 KubeCon + CloudNativeCon + Open Source Summit - 线上峰会 才能参加会议。如果你还没有注册但想加入我们,请到活动注册页面购票注册。

请注意:此日程表自动显示为中国标准时间(UTC +8)。要想看到您选择的时区,请从右侧 「Filter by Date」上方的下拉菜单中选择。日程表可能会有变动。


Virtual
December 9-10
Learn More and Register to Attend

The Sched app allows you to build your schedule but is not a substitute for your event registration. You must be registered for KubeCon + CloudNativeCon + Open Source Summit China 2021 - Virtual to participate in the sessions. If you have not registered but would like to join us, please go to the event registration page to purchase a registration.

Please note: This schedule is automatically displayed in China Standard Time (UTC +8). To see the schedule in your preferred timezone, please select from the drop-down menu to the right, above "Filter by Date." The schedule is subject to change.
Back To Schedule
Friday, December 10 • 11:20 - 11:55
深入研究: 基于 CRI-RM 的中央处理器和非统一内存访问架构亲和性实现人工智能任务加速 | Deep Dive: CRI- RM Based CPU and NUMA Affinity to Achieve AI Task Acceleration - Dekui Wang, Inspur & Jerry Wang, Intel

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
我们与英特尔合作,在 k8s 中使用 CRI-RM 进行节点内资源分配,以加速人工智能培训,其中 CRI-RM 组件是一个用于节点内资源分配的开源项目。对于一些开发者来说,他们尝试基于中央处理器运行人工智能任务。目前,k8s 中的中央处理器和非统一内存访问架构特性只适用于有 qos 保证的普通旧数据,这可能有一些限制:1.中央处理器和拓扑管理器代码集成在 Kubelet 组件中,这对于自定义开发并不容易。2.对于某些人工智能场景,开发人员不想限制普通旧数据的内存,他们不能使用 k8s 的功能。我们提出了一种基于中央处理器的人工智能任务加速方案,可以将人工智能任务的性能提高 50% 以上。通过集成 CRI-RM 组件,我们可以充分利用主机的拓扑信息。这可能有以下优点:1.CRI-RM 可定制开发,并可用于较低版本的 k8s 2. 中央处理器绑定和非统一内存访问架构亲和性对于没有内存限制的普通旧数据也很有效。

We work with Intel to use CRI-RM for in-node resource allocation in k8s to accelerate AI training, where the CRI-RM component is an open source project for resource allocation within node. For some developers, they try to run AI tasks based on CPU. At present, the CPU and NUMA feature in k8s will only work for pods with guaranteed qos, this may have some limitations: 1. CPU and topology manager codes are integrated in Kubelet component, which is not easy for customize development. 2. For some AI scenarios, developers don’t want to limit pod’s memory, they can't use the feature of k8s. We propose a CPU-based AI task acceleration scheme, which can improve the performance of AI tasks by more than 50%. By integrating CRI-RM component, we can make full use of the topology information of the host. This may have these advantages: 1.CRI-RM can be customized for development, and can be used in lower version k8s 2.CPU binding and NUMA affinity can also be effective for pod without memory limit.

Speakers
avatar for Dekui Wang

Dekui Wang

software architect, inspur
毕业于西安电子科技大学,熟悉云计算、AI平台等相关技术,目前主要专注于容器、AI技术以及两者相结合的技术研究。
avatar for Jerry Wang

Jerry Wang

架构师, Intel
毕业于西安电子科技大学,熟悉云计算,UEFI技术,现在英特尔DCAI CESG ESS 部门担任解决方案架构师。曾负责英特尔精选开源云解决方案,KVM优化,高性能计算上云等项目。



Friday December 10, 2021 11:20 - 11:55 CST
Kubecon + CloudNativeCon 演讲厅