Probabilistic Workload Estimation for Server Consolidation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Managing distributed computing systems with uncorrelated and non-deterministic workload patterns is challenging due to the difficulty in accurately estimating combined workloads, leading to underutilized servers and increased costs, as existing methods fail to account for stochastic workload contention.
Innovation Solution
A method that uses quantile-based workload data, normalization, and probabilistic models to estimate combined workloads by specifying a confidence level, allowing for the calculation of workload contention probability and resulting in more accurate consolidation scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If systems are consolidated to reduce costs, then operating costs are reduced, but accurate workload estimation becomes more difficult due to uncorrelated and non-deterministic workload patterns
Solution Approach 1:
The patent transforms workload characterization from deterministic time-series values to probabilistic quantile parameters. By representing workloads as quantile distributions (e.g., 5th, 50th, 95th percentiles) rather than fixed time-series values, the system can accurately estimate combined workloads of uncorrelated systems. This parameter transformation enables precise consolidation planning while maintaining cost efficiency.
2Measurement precision
If sophisticated probabilistic models are used to model combined workloads, then workload estimation accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the complex probabilistic workload modeling into distinct quantile calculations. Instead of modeling entire workload distributions simultaneously, the system calculates specific quantile values (5th, 50th, 95th percentiles) independently using simplified formulas. This segmentation reduces computational complexity while maintaining accuracy, as each quantile can be computed separately using basic statistical operations rather than full probabilistic simulations.
3Measurement precision
If quantile-based workload data is used instead of time-series data, then combined workload estimation for uncorrelated systems improves, but data normalization becomes more complex
Solution Approach 1:
The patent normalizes quantile-based workload data by transforming different quantile measurements into a unified scale. The system applies normalization factors to convert quantile values from different systems into comparable metrics, enabling accurate combined workload estimation. This parameter transformation approach simplifies the normalization process by working with standardized quantile parameters rather than raw time-series data.
Data Source
AI summary
It has been found that a more reasonable estimation of combined workloads can be achieved by enabling the ability to specify the confidence level in which to estimate the workload values. A method, computer readable medium and system are provided for estimating combined system workloads. The method comprises obtaining a set of quantile-based workload data pertaining to a plurality of systems and normalizing the quantile-based workload data to compensate for relative measures between data pertaining to different ones of the plurality of systems. A confidence interval may then be determined and the confidence interval used to determine a contention probability specifying a degree of predicted workload contention between the plurality of systems according to at least one probabilistic model. The contention probability may then be used to combine workloads for the plurality of systems and a result indicative of one or more combined workloads then provided.