Storage Node Scheduler Tree for Real-Time Multi-Tenant IO Isolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing multi-tenant IO quality management methods in elastic block storage systems lack real-time performance and flexibility, leading to insufficient sub-second level IO quality management and inflexible tenant-specific capability recognition.
Innovation Solution
Implementing a traffic scheduling method on storage nodes using a scheduler tree structure that includes a priority scheduler, tenant-level scheduler, and first-in first-out scheduler, based on tenant lists that maintain service capabilities, to allocate and isolate traffic within the storage cluster.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If a control node periodically collects feedback and makes decisions to allocate resources, then resource allocation can be made, but real-time performance is insufficient with intervention time at second level
Solution Approach 1:
The patent extracts the scheduling decision-making function from the control node and relocates it to storage nodes. Each storage node independently determines its own resource allocation based on local feedback and pre-configured weight coefficients, eliminating the need for periodic control node intervention and achieving sub-second real-time performance
Solution Approach 2:
The patent implements local scheduling at each storage node based on its specific workload characteristics and tenant weight coefficients. Each storage node independently adjusts its resource allocation according to local conditions rather than receiving centralized commands, enabling faster and more adaptive resource management
2Adaptability or versatility
If basic real-time IO quality management capabilities are implemented at storage nodes, then some traffic control can be achieved, but flexibility and sub-second level management are insufficient
Solution Approach 1:
The patent segments the scheduling mechanism into three hierarchical levels: priority scheduling for critical traffic, weight-based proportional scheduling for multi-tenant fairness, and FIFO scheduling for standard traffic. This segmented approach provides flexible adaptation to different traffic scenarios while maintaining manageable system complexity through modular design
Solution Approach 2:
The patent implements dynamic scheduling where storage nodes continuously adjust resource allocation based on real-time feedback and configurable weight coefficients. The system transitions from static periodic control to dynamic continuous adjustment, enabling sub-second responsiveness and flexible adaptation to changing workload conditions
3Reliability
If traffic limiting and fixed priority IO scheduling are used, then basic IO quality management is achieved, but tenant-specific performance allocation and isolation are insufficient
Solution Approach 1:
The patent implements a nested scheduling structure where multiple scheduling algorithms (priority scheduling, weight-based proportional scheduling, and FIFO scheduling) are organized in a hierarchical tree structure. Each scheduling layer operates independently but is nested within the overall scheduling framework, enabling effective tenant isolation through multiple levels of control while maintaining organized system complexity
Data Source
AI summary
The present disclosure provides a traffic scheduling method for multiple tenants, a medium and an electronic device. The method includes: receiving, by a storage node in a storage cluster, a tenant list sent by a control node in the storage cluster; determining, by the storage node, a scheduler tree based on the tenant list, where the scheduler tree includes a priority scheduler, a tenant-level scheduler, and a first-in first-out scheduler, the priority scheduler is configured to schedule traffic of the multiple tenants and traffic inside the storage cluster according to business priorities, the tenant-level scheduler is configured to schedule traffic between the multiple tenants, and the first-in first-out scheduler is configured to schedule traffic in and out; and controlling, based on the scheduler tree, traffic in and out of the storage cluster.