Service-Based Workload Management in Multi-Node Database Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for measuring and managing workload in multi-node database systems are limited by their focus on physical components, failing to correlate well with how resources are used by applications, which hinders effective performance analysis and management, particularly in systems like online order entry applications.
Innovation Solution
The approach defines logical abstractions called services, which categorize work across nodes, enabling global and location-transparent performance measurement and management, allowing for real-time performance monitoring and dynamic resource allocation to meet quality of service requirements, such as shifting workload between nodes or instances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If workload information is demarcated by physical components (node, database instance, user session), then measurement precision for physical resources is improved, but adaptability to application-specific performance management deteriorates
Solution Approach 1:
The patent segments workload measurement into two independent dimensions: physical component level (node, instance, session) and service level (application-specific). This allows simultaneous tracking of both physical resource utilization and application performance, resolving the contradiction by enabling independent measurement at each level without interference.
Solution Approach 2:
The patent introduces service identifiers as an intermediary layer that connects physical components to applications. By associating workload information with service identifiers, the system can track application-specific performance while maintaining physical component measurements, acting as a mediator between the two measurement levels.
2Device complexity
If workload information is demarcated by physical components, then device complexity for tracking physical resources is reduced, but information completeness for application performance management deteriorates
Solution Approach 1:
The patent separates tracking complexity into two independent systems: one for physical components and one for services. Each system maintains its own data structures and measurement methods, preventing the complexity of application-specific tracking from overwhelming the physical resource monitoring infrastructure.
Solution Approach 2:
The patent creates a universal service identifier system that can be applied across multiple applications and physical configurations. This single mechanism serves multiple functions: tracking application performance, allocating resources, and generating reports, thereby providing complete application performance information without proportionally increasing overall system complexity.
3Productivity
If resources are allocated based on physical resource utilization, then productivity of resource allocation is improved, but adaptability to application-specific quality of service requirements deteriorates
Solution Approach 1:
The patent implements feedback loops that continuously monitor both physical resource utilization and service-level performance metrics. This dual feedback system enables dynamic resource allocation decisions that respond to both efficiency metrics and quality of service requirements, allowing the system to adjust allocations in real-time based on actual application needs.
Solution Approach 2:
The patent creates dynamic resource allocation that can adapt between different allocation strategies. When physical resource utilization is the primary concern, efficient allocation is applied; when application-specific quality of service requirements become critical, the system dynamically shifts to service-level prioritization, making the allocation mechanism flexible and adaptive.
Data Source
AI summary
Approaches for manage and measuring workload on computer resources by measuring workload in ways that are independent of physical hardware and that correlate to ways in which it is desirable to measure and manage workload for applications and clients. Specifically, logical abstractions of workload, referred to as services, are defined for a cluster of nodes running a multi-node database server of a database. A service is a category of work hosted on a subset of multi-node database servers, such as work that is performed for an online order entry application. The workload is measured and demarcated by the service no matter which node in a single or multi-node system performs the service, providing a global view of performance realized by the system.