SoC Performance Prediction Module Using History Table
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current processor performance control techniques, such as dynamic frequency scaling and dynamic voltage and frequency scaling, fail to accurately reflect software performance factors, leading to increased prediction errors and inefficient power consumption due to reliance on hardware-centric data.
Innovation Solution
A system on chip (SoC) with a performance prediction module that generates control information by referencing a history table of accumulated performance data from previous executions of functions, allowing for more accurate predictive performance control and energy-efficient processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If dynamic frequency scaling or dynamic voltage and frequency scaling techniques are used to control processor performance, then power consumption is reduced, but prediction accuracy deteriorates because hardware-centric data does not reflect software performance factors
Solution Approach 1:
The patent segments performance information into two distinct categories: hardware factors (processor frequency, voltage, temperature) and software factors (application type, function characteristics, execution patterns). By separating these factors and collecting them independently, the system can process each type of information through appropriate measurement and prediction mechanisms, thereby improving overall prediction accuracy while maintaining power efficiency.
Solution Approach 2:
The patent implements a feedback mechanism where actual performance measurements from both hardware and software factors are continuously collected, compared with predicted values, and used to refine future predictions. The performance management module uses this feedback loop to adjust prediction models and control decisions, ensuring that prediction accuracy improves over time while maintaining optimal power consumption levels.
2Loss of energy
If hardware-centric performance data is collected for prediction, then power consumption is reduced through DVFS techniques, but prediction error increases due to lack of software factor information
Solution Approach 1:
The patent creates a universal performance measurement framework that simultaneously handles both hardware metrics (frequency, voltage, temperature) and software metrics (application characteristics, function types, execution patterns). This multi-functional approach allows the system to gather comprehensive performance information through a unified collection mechanism, improving prediction reliability without requiring separate specialized systems for each factor type.
Solution Approach 2:
The patent performs preliminary classification and collection of software factors before performance prediction occurs. By identifying application types, function characteristics, and execution patterns in advance, the system prepares accurate prediction inputs that reflect actual software behavior. This preliminary action ensures that when prediction and control decisions are made, both hardware and software factors are already measured and integrated, enhancing prediction reliability.
3Measurement precision
If comprehensive performance information including software factors is collected, then prediction accuracy is improved, but system complexity increases
Solution Approach 1:
The patent introduces a performance management module as an intermediary that coordinates between hardware sensors, software analysis components, and the prediction algorithm. This mediator consolidates diverse performance information from multiple sources, standardizes the data format, and presents unified inputs to the prediction mechanism. By acting as an intermediary, the system manages complexity through centralized coordination rather than requiring direct integration of all components.
Solution Approach 2:
The patent creates simplified representations (copies) of complex software factors by categorizing applications into types and functions into standardized characteristics. Instead of processing raw, detailed software information directly, the system uses these abstracted copies that capture essential performance-relevant features. This copying approach maintains prediction accuracy by preserving key behavioral patterns while reducing the complexity of data processing.
Data Source
AI summary
A system on chip includes a processor configured to execute an application by using a shared library, the application being supported by an operating system, a performance prediction module configured to generate control information of the processor for executing a function by the processor and predictive performance information of the processor that is predicted during control according to the control information when the function included in the shared library is executed the control information referencing a history corresponding to the function from a history table, and a performance management module configured to generate a control signal for performance control of the processor based on the control information, wherein the history is an accumulation of performance information of the processor measured by executing in advance the function at least once.