Liang Yan
Coreweave, Sr. Software Engineer
Actions
Liang Yan is a senior software engineer at Coreweave, specializing in AI Infra, heterogeneous architecture acceleration and distributed machine learning systems from the cloud base. He collaborates closely with upstream communities and leading vendors like NVIDIA, AMD and ARM, delivering creative solutions to the teams and customers. He has also been passionate about open source and Linux for over a decade. Liang has delivered insightful presentations at prestigious conferences, including NVIDIA GTC, AI_DEV, KubeCon, LPC, LSFMM, KVM Forum, etc.
Optimize the Ray schedule and autoscaling on Kubernetes Cloud: A Heterogeneous Task Perspective
Notably, K8s has witnessed a remarkable surge in adoption within the machine learning domain, benefiting from its fantastic container orchestration capability. Ray with KubeRay is emerging as a prominent player. Ray is also a distributed framework known for scaling workloads to a cluster. However, both mechanisms rely on limited metric thresholds and uniform node scaling, which may not prove entirely efficient for certain scenarios.
In this session, a new scheduling strategy will be presented, which takes the task type into consideration. Moreover, it will explore the flexibility to autoscale K8s nodes from different VM flavors based on resource request calculation. This approach holds the potential to enhance both the load-balance and cost-effectiveness of running the Ray cluster on VM-based K8s from the public cloud.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top