Core Assignment Optimizer for Heterogeneous Processing Engines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing systems face complexity in workload balancing due to heterogeneous processing engines with unique power/performance profiles, leading to inefficiencies in power consumption and performance optimization, as they often lack comprehensive understanding of hardware characteristics and rely on user preferences or basic workload classifications.
Innovation Solution
A core assignment optimizer that selects processing engines based on static and heuristic profiling, considering user preferences, system capabilities, power/performance profiles, processing engine usage, thermal characteristics, and resource availability to optimize workload distribution across heterogeneous cores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If workload is assigned to heterogeneous processing engines based on basic workload classifications, then workload distribution is achieved, but power efficiency and performance optimization are insufficient
Solution Approach 1:
The system implements feedback mechanisms by continuously monitoring workload characteristics, core performance metrics, and power consumption data. The workload classifier and core selector use this feedback to dynamically adjust workload assignments, matching workloads to cores that optimize both performance and power efficiency based on real-time system state.
Solution Approach 2:
The system changes operational parameters by adjusting workload-to-core assignments based on multiple variables including workload type, core power/performance profiles, thermal states, and power budget constraints. This dynamic parameter adjustment enables the system to optimize power efficiency while maintaining productivity across heterogeneous cores.
2Productivity
If more heterogeneous processing engines are added to computing systems, then processing capability increases, but system complexity and power management difficulty increase
Solution Approach 1:
The patent introduces intermediary components including a workload classifier, core selector, and power management module that mediate between the diverse heterogeneous cores and incoming workloads. These intermediaries abstract the complexity of managing multiple heterogeneous processing engines, providing a unified interface for workload submission while handling the complex task of matching workloads to appropriate cores based on their unique characteristics.
Solution Approach 2:
The system implements universal workload classification and core selection mechanisms that can handle any type of workload across any heterogeneous core. The workload classifier and core selector are designed to be agnostic to specific core architectures, enabling the system to manage diverse processing engines through a unified, multi-functional framework that reduces overall system complexity.
3Ease of operation
If user preferences for performance or power are allowed, then user control is provided, but optimal workload-to-core assignment requires deeper hardware understanding
Solution Approach 1:
The system enables self-service operation by automatically interpreting user preferences (performance-oriented or power-efficient mode) and autonomously performing workload classification, core selection, and assignment. The power management module and core selector work together to translate high-level user preferences into detailed hardware-level decisions, eliminating the need for users to understand complex hardware characteristics while still providing granular control over power and performance settings.
Data Source
AI summary
Embodiments described herein may include apparatus, systems, techniques, and/or processes that are directed to techniques for workload balancing in a computing system with heterogeneous processing engines. A core assignment optimizer selects one or more of the heterogeneous processing engines based on static and heuristic profiling. Static factors considered may include user preferences, system capabilities including maximum power consumption and thermals, the workload type and the unique power/performance profile of each heterogeneous processing engine and the like. Dynamic or heuristic factors may include processing engine usage over time, current thermal characteristics of the system, current system workload including resource availability, and the like.