QoS Deployment Priorities for Multi-Tenant Cluster Capacity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing container orchestration systems in multi-tenant clusters struggle to efficiently allocate resources based on quality of service (QoS) levels, leading to inefficiencies in resource utilization and potential resource starvation among different tenant services.
Innovation Solution
Implementing deployment objects associated with different QoS levels, each with a combined priority value, to manage resource allocation and ensure critical services receive adequate resources by preempting or evicting lower-priority services when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If resources are allocated without QoS-based priority levels, then resource allocation is simple, but critical services may experience resource starvation and service reliability deteriorates
Solution Approach 1:
The patent segments resources into different QoS levels (guaranteed, burstable, best-effort) with distinct priority values. This segmentation allows the cluster to differentiate between critical and non-critical services, ensuring reliable resource allocation to high-priority services while maintaining manageable complexity through structured categorization.
Solution Approach 2:
The patent applies local quality by assigning different quality attributes to different resource allocations. Each QoS level has specific characteristics (guaranteed capacity, burstable capacity, or best-effort) that are locally optimized for different service needs, enabling reliable service delivery to critical applications without uniformly treating all services equally.
2Productivity
If deployment objects with combined priority values are used, then resource allocation efficiency improves, but system complexity increases
Solution Approach 1:
The patent merges service priority and QoS level priority into a single combined priority value within the deployment object. This merging simplifies the scheduling decision-making process by providing a unified priority metric, thereby improving resource allocation efficiency while containing complexity through consolidation rather than multiplication of separate evaluation criteria.
Solution Approach 2:
The deployment object serves multiple functions: it specifies service details, defines QoS level, assigns combined priority value, and establishes deployment quotas. This multi-functionality reduces the need for separate configuration mechanisms, improving allocation efficiency while managing complexity by consolidating multiple concerns into a single universal object type.
3Stability of the object's composition
If deployment quotas are enforced for each QoS level, then service stability is maintained, but deployment flexibility is reduced
Solution Approach 1:
The patent implements dynamic quota management where deployment quotas for each QoS level can be adjusted based on cluster state and service needs. The system dynamically determines whether to enforce quotas strictly or allow flexibility, enabling stability through quota enforcement when needed while maintaining adaptability when resource availability permits deviation from quota limits.
Data Source
AI summary
Instances of a service are deployed at different quality of service (QOS) levels associated with different instance priorities. A manifest for a service specifies a first QoS level associated with a first QoS level priority value. A first deployment object is created for deploying instances of the service at the first QoS level, and is associated with a first combined priority value determined based on the priority of the service and the first Qos level priority value. The first deployment object is further associated with a deployment quota associated with deployment of the service at the first QoS level. An instance of the service is deployed using the first deployment object when the instances of the service currently deployed in the cluster satisfy a predetermined relationship with the deployment quota associated with the first QoS level.