Linear-Time Shared Footprint Quantification for Multi-Threaded Cache Optimization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for quantifying digital data sharing in multi-threaded executions are inefficient, particularly due to high asymptotic complexity and resource usage, which hinders effective analysis and optimization of run-time performance and cache scaling.

Innovation Solution

A system and method that calculates shared footprint values and average shared footprint values using a linear-time process, allowing for the quantification of digital data sharing across multiple threads and cache sizes, enabling precise measurement and optimization of thread-core mapping and cache performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional methods are used to quantify digital data sharing in multi-threaded executions, then measurement capability is provided, but computational efficiency deteriorates due to high asymptotic complexity and resource usage

Engineering Contradiction:

Improvedata sharing quantificationVSAvoidcomputational efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent changes the computational parameters by using a linear-time process that calculates shared footprint values through a single pass over the execution trace, transforming the asymptotic complexity from super-linear to linear time and space, thereby resolving the contradiction between measurement precision and computational efficiency

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent segments the measurement process into distinct components: identifying shared data, calculating individual shared footprint values for different cache sizes, and computing average shared footprint values. This segmentation allows each component to be processed efficiently in a linear-time manner

Inventive Principle:

Principle #1Segmentation

2Manufacturing precision

If detailed analysis of data sharing is performed, then optimization accuracy is improved, but resource consumption increases

Engineering Contradiction:

Improveoptimization accuracyVSAvoidresource consumption

Core Design Contradiction:

Manufacturing precisionVSQuantity of substance

Solution Approach 1:

The patent performs preliminary action by calculating all shared footprint values and average shared footprint values in a single pass over the execution trace before optimization begins. This preliminary calculation provides accurate measurement data that guides subsequent optimization decisions, improving optimization accuracy while avoiding repeated expensive analyses

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS9535704B2System and method to quantify digital data sharing in a multi-threaded execution

Publication Date: 2017.01.03 UNIVERSITY OF ROCHESTER

AI summary

A method to quantify a plurality of digital data sharing in a multi-threaded execution includes the steps of: providing at least one processor; providing a computer readable non-transitory storage medium including a computer readable multi-threaded executable code and a computer readable executable code to calculate a plurality of shared footprint values and an average shared footprint value; running the multi-threaded executable code on the at least one computer processor; running the computer readable executable code configured to calculate a plurality of shared footprint values and an average shared footprint value; calculating a plurality of shared footprint values by use of a linear-time process for a corresponding plurality of executable windows in time; and calculating and saving an average shared footprint value based on the plurality of shared footprint values to quantify by a metric the data sharing by the multi-threaded execution. A system to perform the method is also described.