Decentralized Resource Scheduling via Workload Billboard
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Centralized resource scheduling in cloud computing faces challenges such as high computational requirements, limited scalability, and inefficient resource utilization, particularly in large clusters, where pre-defined resource definitions may not be effective across different system architectures and can lead to increased recovery times during disasters.
Innovation Solution
A decentralized resource scheduling approach where each computer host in a cluster manages its own resources through an inquiry-handshake model, using a proxy to collect resource usage information and publish workload details to a shared workload billboard, allowing for dynamic resource allocation and scheduling without a centralized scheduler.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized resource scheduling is used, then resource allocation can be coordinated across the cluster, but computational requirements increase and scalability is limited
Solution Approach 1:
The patent divides the centralized scheduling function into distributed scheduling units at each host. Each scheduling unit independently manages resource allocation for its local host, eliminating the need for a single centralized scheduler. This segmentation reduces computational burden on any single node while maintaining coordinated resource allocation across the cluster through peer-to-peer communication.
2Productivity
If centralized resource scheduling is used, then resource coordination is achieved, but scalability to large clusters is limited
Solution Approach 1:
By segmenting the scheduling function across multiple independent scheduling units, the system can scale to large clusters without a single point of failure or bottleneck. Each host operates autonomously with its own scheduling unit, allowing the cluster to grow dynamically without requiring proportional increases in centralized computational resources.
Solution Approach 2:
The patent transitions from a vertical hierarchical architecture (centralized scheduler at top) to a horizontal peer-to-peer architecture where scheduling units operate at the same level. This dimensional change enables better scalability as each node operates independently while still contributing to overall cluster resource coordination.
3Ease of operation
If pre-defined resource definitions are used, then resource allocation is simplified, but effectiveness across different system architectures is reduced
Solution Approach 1:
The patent implements dynamic resource definitions that adapt to different system architectures and workload characteristics. Instead of using fixed pre-defined resource definitions, the scheduling units can adjust resource parameters and allocation strategies based on real-time conditions and architectural differences, maintaining both simplicity and cross-architecture effectiveness.
Solution Approach 2:
The system allows resource definition parameters to be changed and customized for different architectures. Each scheduling unit can modify resource allocation parameters to match the specific characteristics of its host system, enabling effective resource management across diverse architectures while maintaining operational simplicity through standardized interfaces.
4Productivity
If centralized scheduling is used, then resource management is unified, but recovery time during disasters increases
Solution Approach 1:
By distributing the scheduling function across multiple independent units rather than relying on a single centralized scheduler, the system eliminates single points of failure. During disasters, individual scheduling units can continue operating independently or be replaced without affecting the entire cluster, significantly reducing recovery time while maintaining unified resource management through coordinated operation of remaining units.
Solution Approach 2:
The distributed architecture provides built-in redundancy and fault tolerance. Each host has its own scheduling unit that can independently handle failures, providing a cushion against disasters before they impact the entire system. This prior cushioning through distribution ensures faster recovery compared to centralized systems that must wait for the central scheduler to recover.
Data Source
AI summary
Methods, apparatus, computer program products for resource scheduling are provided. The method comprises: receiving a workload request; publishing the information of the workload to a workload billboard accessible to a plurality of computer hosts, each of the plurality of computer hosts being associated with a corresponding proxy configured to manage the resource scheduling of the computer host; receiving a request to schedule at least a portion of the workload from a proxy; and sending the portion of the workload to the computer host associated with the proxy.