Performance Impact Factor for Server Load Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analyzing performance data from large server farms is impractical due to the vast amount of data generated by frequent sampling of performance counters, which is often stripped of meaningful information by abstractions like averages and standard deviations, and visual analysis of graphs becomes labor-intensive and inefficient.
Innovation Solution
Calculating a performance impact factor using a weighted sum of performance counter values, where weights are assigned based on the relative significance of different utilization ranges, allowing for the identification of machines experiencing significant performance impacts through a metric that reflects the actual or estimated performance impact.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If performance counters are sampled frequently to capture short-duration performance spikes, then detection precision is improved, but data volume increases making storage and analysis impractical
Solution Approach 1:
The patent extracts only the essential performance impact information from raw counter data by calculating a performance impact factor that captures the essence of performance degradation without retaining all raw data points. This selective extraction maintains detection precision while dramatically reducing data volume for storage and analysis.
Solution Approach 2:
The patent transforms raw performance counter values into a derived parameter (performance impact factor) that changes based on the severity and duration of performance impacts. This parameter transformation condenses high-frequency data into a manageable format that preserves critical performance information.
2Quantity of substance
If abstractions like averages are applied to simplify performance data analysis, then data volume is reduced, but meaningful information is lost
Solution Approach 1:
The patent applies local quality by treating different performance counter values differently based on their significance. Instead of uniform averaging, the performance impact factor calculation applies weighted treatment to counter values, giving appropriate emphasis to critical performance indicators while downweighting less significant data, thereby preserving meaningful information while reducing data volume.
3Loss of information
If standard deviation is calculated to describe data distribution, then some information about variability is retained, but it is ineffective for non-normal distributions common in performance data
Solution Approach 1:
The patent changes the parameter used to describe performance data distribution from traditional statistical measures (standard deviation) to a performance impact factor that is specifically designed to capture performance degradation patterns. This parameter transformation provides accurate distribution description for the non-normal distributions typical of performance data.
4Ease of operation
If graphs are generated for each machine to visualize performance data, then visual analysis is enabled, but labor intensity increases making analysis impractical for large server farms
Solution Approach 1:
The patent extracts the essential performance impact information into a single metric (performance impact factor) that can be directly analyzed without requiring visual inspection of graphs. This extraction enables efficient analysis of large numbers of machines by presenting condensed, actionable information in a format that does not require time-consuming visual interpretation.
Data Source
AI summary
Analysis may be made of the amount that a load on a machine impacts the machine's performance. Performance counters on the machine record raw statistical data, such as a given resource's current utilization. The values of these counters may be captured. A n-bin histogram may be created that shows how many of the captured performance counter values occur within various ranges, such as 0-10% utilization, 10-20%, etc. A weight may be assigned to each bin. A weighted sum of the bins may be calculated by multiplying the number of occurrences in each bin by the bin's weight, and adding the products together. The weights may be chosen to reflect the relative amounts that particular performance counter values impact the overall performance of a machine. Thus, a metric that represents performance impact may be calculated based on the weighted sum.