Statistical Representative Data Pooling for Cloud Resource Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In cloud computing and client-server environments, maintaining historical resource usage data for load balancing and resource management is resource-intensive, consuming significant compute and storage resources, and can lead to inefficient VM placement due to the need for precise historical data.
Innovation Solution
The approach involves partitioning historical resource usage data into pools, computing and storing statistical representatives for each pool, which summarize resource usage statistics, allowing for reduced data storage and computation while maintaining accurate historical data for longer durations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If historical resource usage data is stored in full detail for load balancing and resource management, then decision accuracy is improved, but memory consumption and computational overhead increase significantly
Solution Approach 1:
The patent segments historical resource usage data into fixed-size pools (e.g., pools of 10 data points). Each pool is processed independently to compute statistical representatives, reducing the memory burden of storing and processing entire historical datasets while maintaining sufficient accuracy for load balancing decisions.
Solution Approach 2:
Instead of storing complete historical resource usage data, the patent creates simplified copies in the form of statistical representatives (mean, standard deviation, minimum, maximum values) that capture the essential characteristics of each data pool. These compact representations consume significantly less memory while preserving the necessary information for effective resource management decisions.
2Reliability
If complete historical resource usage data is maintained for accurate VM placement decisions, then resource management quality is improved, but computational overhead increases
Solution Approach 1:
The patent divides historical data into fixed-size pools and processes each pool independently to compute statistical representatives. This segmentation reduces the computational complexity of analyzing entire historical datasets while maintaining the reliability needed for accurate VM placement and load balancing decisions.
Solution Approach 2:
The patent transforms raw historical resource usage data into transformed parameters (statistical representatives including mean, standard deviation, minimum, and maximum values). This parameter transformation reduces computational overhead by working with aggregated statistics rather than individual data points, while preserving the essential information needed for reliable resource management.
3Quantity of substance
If statistical representatives are computed for fixed-size pools of historical data, then memory efficiency is improved, but data granularity is reduced
Solution Approach 1:
The patent segments historical data into fixed-size pools (e.g., pools of 10 data points) and computes statistical representatives for each pool. This segmentation achieves memory efficiency by limiting the storage requirements to a manageable number of pools and their representatives, while the fixed pool size preserves sufficient data granularity for effective load balancing and resource management decisions.
Data Source
AI summary
Techniques for representative-based approach to store historical resource usage data are disclosed. In one embodiment, a management node may include a statistical representative computational unit to obtain historical resource usage data associated with a workload running on a host, divide the historical resource usage data into a plurality of pools with each pool comprising a predefined number of resource usage statistics, and determine a statistical representative corresponding to each of the pools. Each statistical representative may represent the predefined number of resource usage statistics in a corresponding one of the pools. Further, the management node may include a monitoring and managing unit to monitor and/or manage the workload using the statistical representatives.