Grid Resource Allocation Using Predicted Compute Availability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems do not provide clients with the calculation capability of grid computing effectively.
Innovation Solution
A management system that allocates arithmetic devices in grid computing based on job data calculation type, execution condition, and delivery date, ensuring appropriate allocation and timely completion of processing tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If arithmetic devices are allocated based on simple availability without considering calculation type and performance matching, then resource allocation is simple and quick, but calculation efficiency and task completion quality deteriorate
Solution Approach 1:
The management server performs preliminary classification of arithmetic devices by calculation type (CPU, GPU, FPGA) and pre-stores their performance information in a database. When job data arrives, the system quickly matches the required calculation type with available devices, avoiding complex real-time analysis and enabling efficient allocation based on pre-established performance profiles
Solution Approach 2:
The system dynamically adjusts device allocation by continuously monitoring the operational status of arithmetic devices and updating availability in real-time. The matching process adapts to changing conditions by selecting from currently available devices that match the calculation type requirements, ensuring optimal resource utilization while maintaining allocation efficiency
2Loss of time
If the system waits for current resource availability without prediction, then allocation decisions are simple, but task completion timing deteriorates when delivery dates are critical
Solution Approach 1:
The system performs preliminary prediction of future resource availability by analyzing current device status and projected operational patterns. This advance prediction enables the matching process to identify devices that will be available by the delivery date, preventing time-critical tasks from being delayed due to temporary unavailability of suitable arithmetic devices
Solution Approach 2:
The system continuously monitors the actual operational status of arithmetic devices and compares it with predicted availability. This feedback mechanism allows the system to refine its predictions and adjust allocations dynamically, ensuring that delivery date requirements are met while adapting to real-world variations in device performance and availability
3Ease of operation
If grid computing resources are not systematically managed and allocated, then system simplicity is maintained, but clients cannot effectively access calculation capabilities
Solution Approach 1:
The management server acts as an intermediary between clients and the distributed arithmetic devices. It receives job data from clients, automatically matches requirements with available devices based on calculation type and performance, and coordinates execution. This intermediary layer shields clients from the complexity of resource management while providing efficient access to grid computing capabilities
Solution Approach 2:
The management server performs multiple functions including resource discovery, performance evaluation, job matching, task scheduling, and result aggregation. This multi-functional approach consolidates complex grid computing management into a single accessible interface, allowing clients to obtain calculation capabilities without dealing with underlying system complexity
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
In job acceptance processing, a controller (505) accepts job data requested for calculation by a client. In prediction processing, the controller (505) predicts a temporal change in a calculation capability available in grid computing processing of each of a plurality of arithmetic devices (105). In matching processing, the controller (505) allocates an arithmetic device (105) that is available to the job data accepted in the job acceptance processing on the basis of a result of the prediction processing.