Dynamic Workload Adjustment via Saturation Metric
Solution Overview
Problem
Existing data storage systems lack an accurate and linear method to estimate unused processing capacity, known as 'headroom,' which is essential for dynamic workload adjustment and efficient resource utilization.
Innovation Solution
A saturation model is developed that calculates a saturation metric scaling linearly with IOPS rates, allowing for workload adjustments by determining if the metric is above or below thresholds, and applying additional processing to mode-switching metrics for core allocation among different processing tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a saturation metric is calculated to estimate headroom for dynamic workload adjustment, then resource utilization efficiency is improved, but the system requires additional processing of mode-switching metrics and core allocation patterns which increases system complexity
Solution Approach 1:
The patent introduces a saturation metric as an intermediary that mediates between the complex internal system state (mode-switching metrics, core allocation patterns) and the external workload adjustment decision. This saturation metric simplifies the complex internal state into a single value that can be used for headroom estimation, thereby improving resource utilization efficiency without requiring the external system to directly process all the complex internal parameters.
Solution Approach 2:
The patent transforms multiple system parameters (mode-switching metrics, core allocation patterns, IOPS rates) into a single saturation metric parameter. This parameter transformation allows the system to estimate headroom using a simplified parameter rather than directly processing multiple complex parameters, resolving the contradiction between improved resource utilization and reduced system complexity.
2Measurement precision
If the saturation metric is calculated based on IOPS rate relative to maximum IOPS rate, then headroom estimation accuracy is improved, but the measurement and detection of saturation value becomes more difficult due to the need to monitor and compare multiple threshold values
Solution Approach 1:
The patent segments the saturation metric calculation into distinct components: IOPS rate measurement, maximum IOPS rate determination, and threshold comparison (high threshold and low threshold). This segmentation allows for precise headroom estimation by systematically processing each component separately, making the overall measurement process more manageable and less difficult despite the multiple thresholds involved.
Data Source
AI summary
A method of dynamically adjusting workload of a data storage system includes, while processing a first workload, calculating a saturation value of a saturation metric that scales substantially linearly with an I/O per second (IOPS) rate relative to a maximum IOPS rate of the system, determining that the saturation value is one of (1) above a high threshold and (2) below a low threshold, and performing a workload adjustment operation that establishes a second workload by (1) subtracting from the first workload based on the saturation value being above the high threshold, and (2) adding to the first workload based on the saturation value being below the low threshold, then subsequently processing the second workload.