Decoupled Pod and GPU Node Allocation in Kubernetes Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current resource allocation methods in machine learning scenarios, such as Kubernetes clusters, have low resource utilization rates as they only allocate nodes with GPU resources to pods, failing to optimize the use of available resources across different nodes.
Innovation Solution
A method and apparatus for resource allocation that creates a pod for a target task, acquires GPU resource requirements, and allocates two target nodes: one where the GPU resource is located and another where the pod is allocated, allowing decoupling of node and GPU resource allocation, thereby improving resource utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If only nodes with GPU resources are allocated to pods, then GPU resource allocation is ensured, but resource utilization rate is low
Solution Approach 1:
The patent segments the allocation of nodes and GPU resources by introducing two distinct node types: first target nodes for pod placement and second target nodes for GPU resource location. This segmentation allows independent optimization of pod scheduling and GPU allocation, enabling non-GPU nodes to host pods while GPU resources are allocated from dedicated GPU nodes, thereby improving overall resource utilization while ensuring GPU availability.
2Device complexity
If single node allocation is used, then allocation simplicity is maintained, but resource utilization and efficiency are limited
Solution Approach 1:
The patent transitions from a single-dimension allocation model (one node for both pod and GPU) to a two-dimension allocation model (separate first target nodes for pods and second target nodes for GPUs). This dimensional expansion enables the system to utilize non-GPU nodes for pod hosting while dedicating GPU nodes for computation, effectively adding a new degree of freedom to resource allocation and improving utilization without significantly increasing complexity.
Data Source
AI summary
Provided is a resource allocation method, an electronic device and a storage medium, relating to the field of computer technology, and in particular to fields of resource management, task allocation and the like in computer technology. The resource allocation method includes: creating a pod for a target task; acquiring Graphics Processing Unit (GPU) resource requirement information of the target task; acquiring available node information of a target cluster and available GPU resource information of the target cluster; and allocating, based on the available node information and the available GPU resource information, first and second target nodes satisfying the GPU resource requirement information to the pod, where the first target node is a node where a target GPU resource allocated to the pod is located, and the second target node is a node where the pod allocated to the pod is located.