Sampling-Based Performance Metric Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calculating performance metrics in enterprise computing systems over various historical time periods are neither scalable nor cost-efficient due to the need for storing all historical performance-related data points, which is impractical given the high velocity of transactions in current computing systems.
Innovation Solution
Generating a representative sample of data points using algorithms like Vitter's Algorithm R, which exhibits a mathematical distribution similar to the full set of data points, and calculating metrics objects that describe these distributions, allowing for efficient storage and calculation of performance metrics for different time periods and sub-periods without storing large amounts of data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all historical performance-related data points are stored to calculate performance metrics over various time periods, then measurement precision is improved, but loss of substance worsens due to excessive memory consumption
Solution Approach 1:
The patent creates a representative sample (copy) of the full data set that captures the essential statistical characteristics. Instead of storing all historical data points, a smaller sample data set is generated using algorithms like Vitter's Algorithm R, which produces a sample with the same mathematical distribution properties. This copy enables performance metric calculation while consuming minimal memory resources.
Solution Approach 2:
The patent transforms the raw data into statistical parameters (mean, variance, skewness, kurtosis) that characterize the data distribution. By storing and manipulating these parameters instead of individual data points, the system maintains measurement precision for performance metrics while dramatically reducing memory consumption. The parameters capture the essential information needed for accurate metric calculation.
2Reliability
If all historical data points are stored to enable performance metric calculation, then reliability is improved, but device complexity worsens due to storage infrastructure requirements
Solution Approach 1:
A simplified copy of the data (representative sample) is created that retains the statistical properties needed for reliable performance monitoring. This sample can be generated and stored with minimal infrastructure, eliminating the need for complex large-scale storage systems while maintaining monitoring reliability through statistical representativeness.
Solution Approach 2:
The essential information (statistical parameters characterizing data distribution) is extracted from the full data set. Only these extracted parameters need to be stored, rather than the complete historical data. This extraction simplifies the storage requirements while preserving the reliability needed for performance metric calculation.
3Productivity
If representative samples are generated using algorithms like Vitter's Algorithm R, then productivity is improved through efficient storage and calculation, but device complexity worsens due to algorithm implementation requirements
Solution Approach 1:
The representative sample is generated in advance using algorithms like Vitter's Algorithm R, before performance metrics need to be calculated. This preliminary sampling action captures the statistical characteristics of the data, enabling efficient subsequent metric calculations without requiring complex real-time processing. The sampling algorithm is executed once per data collection period, improving overall productivity.
Data Source
AI summary
Systems and methods include determination of a first data analysis period, determination of a first plurality of sets of parameter values, each of the first plurality of sets of parameter values being associated with a respective time period within the first data analysis period and describing a statistical distribution of data points associated with the respective time period, determination of a statistical distribution associated with the first data analysis period based on the first plurality of sets of parameter values, determination of a system condition based on the statistical distribution associated with the first data analysis period, and initiation of an action based on the determined system condition.