ACC Process Migration for Thermal Load Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-performance computing environments using pooled accelerator cards (ACCs), heat generated by one ACC can affect neighboring cards, leading to unstable operation due to reduced cooling efficiency and elevated temperatures.
Innovation Solution
A system that monitors operating temperatures and processing loads of multiple ACCs, migrating processes from overheated cards to cooler ones, thereby maintaining stable operation by optimizing resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple ACCs are pooled together to increase processing capacity, then productivity is improved, but temperature rises due to heat concentration
Solution Approach 1:
The patent implements dynamic process migration between ACCs based on real-time temperature monitoring. When an ACC's temperature exceeds a threshold, its processes are automatically migrated to other ACCs, creating a dynamic load-balancing system that adapts to thermal conditions and prevents heat concentration while maintaining high productivity
Solution Approach 2:
The system changes the operational parameters of ACCs by monitoring temperature and processing load, and dynamically reallocating processes based on these parameter changes. This allows the system to optimize the balance between productivity and temperature by adjusting process distribution in response to real-time conditions
2Productivity
If processes are concentrated on fewer ACCs to maximize utilization, then productivity is improved, but reliability deteriorates due to thermal instability
Solution Approach 1:
The patent implements a feedback mechanism that continuously monitors temperature and processing load of each ACC, and uses this feedback to dynamically migrate processes. This closed-loop control ensures that productivity is maintained by keeping ACCs actively utilized while reliability is preserved by preventing thermal instability through automatic process redistribution when temperature thresholds are exceeded
3Temperature
If processes are migrated frequently to balance load, then temperature is controlled, but device complexity increases
Solution Approach 1:
The patent applies partial action by implementing process migration only when necessary - specifically when temperature exceeds a predefined threshold or load balance requirements are not met. This selective migration approach controls temperature effectively while avoiding unnecessary migration operations that would increase device complexity and overhead
Data Source
AI summary
A computer-readable recording medium has stored therein a program for causing a computer to execute an information process including: obtaining an operating temperature and a processing load of each of a plurality of process executing devices; and migrating a process assigned to a first process executing device to a second process executing device, the first process executing device and the second process executing device being included in the plurality of process executing devices, the first process executing device having the operating temperature equal to or higher than a first given value and the processing load lower than a second given value, the second process executing device having the operating temperature lower than a third given value equal to or lower than the first given value.