Container Resource Isolation for Low-Latency Specialized Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing container orchestration systems fail to efficiently manage and schedule specialized workloads with specific device and resource requirements, leading to issues like increased latency and resource interference due to OS sharing, especially for virtualized radio access networks (vRANs).
Innovation Solution
Implementing device plugins and scheduler plugins to advertise and manage isolated resources, such as CPU cores and network interfaces, and configuring devices assigned to containers, while using custom runtimes to ensure isolation and resource availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing container orchestration systems are used to manage specialized workloads, then general resource scheduling is maintained, but resource utilization and operational efficiency deteriorate due to inability to meet specific device requirements
Solution Approach 1:
The patent segments CPU resources into isolated CPU cores that are dedicated to specific workloads, separating them from the general CPU pool. This segmentation allows specialized workloads like vRAN to have guaranteed resource allocation while maintaining overall system productivity through efficient resource partitioning and management.
2Loss of time
If OS and other processes share CPU with multiple workloads, then resource sharing efficiency is maintained, but processing latency increases due to resource interference
Solution Approach 1:
The patent extracts isolated CPU cores from the general OS-scheduled CPU pool and dedicates them exclusively to specific workloads. This extraction eliminates OS scheduling overhead and resource interference for time-sensitive operations, reducing processing latency while maintaining overall resource sharing efficiency through proper allocation management.
3Loss of time
If isolated CPU cores are used for low-latency workloads, then processing latency is reduced, but resource availability for general scheduling decreases
Solution Approach 1:
The patent implements dynamic resource management where isolated CPU cores can be allocated and deallocated based on workload requirements. The system dynamically adjusts resource availability by managing the pool of isolated cores, ensuring low-latency performance for specialized workloads while maintaining sufficient resource availability for general scheduling through flexible allocation policies.
4Reliability
If device plugins and scheduler plugins are implemented to manage isolated resources, then resource isolation and availability are improved, but system complexity increases
Solution Approach 1:
The patent introduces device plugins as intermediary components that bridge the gap between the orchestration system and physical devices. These plugins handle the complexity of resource isolation and device management, providing standardized interfaces that simplify the orchestration system while ensuring reliable resource isolation and availability for containerized workloads.
Data Source
AI summary
Techniques are disclosed for managing resources in a computing network comprising a plurality of computing nodes and an orchestrator. Resources are identified that are available for use by workloads to be executed in the computing network. The resources are partitioned into resource classes and exposed to the computing nodes. When a requested resource is reserved for exclusive use, the partitioning of the resources is updated.