Container Scheduler Partitioning for Latency Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional container-orchestration systems face inefficiencies in selecting optimal resources for running workloads due to insufficient information about host system resources, leading to suboptimal performance, especially in latency-sensitive applications like 5G Telco computer architectures.
Innovation Solution
Implementing schedulable partitions and hardware-supported partitioning technologies like NUMA, where resources are partitioned into subsets that can be efficiently managed and allocated by an agent and scheduler, ensuring optimal resource utilization and selection for workloads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional resource selection methods are used in container-orchestration systems, then the system can operate with simple scheduling logic, but the throughput and bandwidth are insufficient due to lack of information about host system resources
Solution Approach 1:
The host system resources are segmented into distinct schedulable partitions (e.g., NUMA nodes, storage resources, compute resources) that can be independently identified and selected by the scheduler. This segmentation enables the scheduler to make informed decisions about which specific resource partitions to allocate to workloads, thereby improving throughput and bandwidth without requiring complete system-wide information.
Solution Approach 2:
An intermediary mechanism is introduced between the scheduler and host system resources that provides targeted information about available resource partitions. This intermediary layer enables the scheduler to access necessary information about host system resources without requiring the scheduler to have complete knowledge of the entire system, thus improving productivity while managing information requirements efficiently.
2Speed
If conventional resource selection methods are used, then the system structure remains simple, but latency performance deteriorates due to inability to select optimal local resources
Solution Approach 1:
The system applies local quality by enabling the scheduler to identify and select local resource partitions (such as local storage resources associated with specific hardware resources) rather than treating all resources uniformly. This allows workloads to be executed on optimally located resources, reducing latency for latency-sensitive applications while maintaining a manageable system structure through targeted rather than comprehensive resource modeling.
3Productivity
If resources are not partitioned into schedulable partitions, then the scheduler operates with simple logic, but resource utilization efficiency decreases due to suboptimal resource selection
Solution Approach 1:
Resources are segmented into schedulable partitions that represent meaningful units of allocation (such as NUMA nodes or storage resources). This segmentation enables the scheduler to efficiently manage and allocate resources without requiring complex fine-grained control, improving resource utilization efficiency while keeping the management complexity at an acceptable level through standardized partition interfaces.
Data Source
AI summary
Techniques of scheduling workload(s) on partitioned resources of host systems are described. The techniques can be used, for example, in a container-orchestration system. One technique includes retrieving information characterizing at least one schedulable partition and determining an availability and a suitability of one or more of the schedulable partition(s) for executing a workload in view of the information. Each of the schedulable partition(s) includes resources of one or more host systems. The technique also includes selecting one or more of the schedulable partition(s) to execute the workload in view of the availability and the suitability.