Storage Array Hardware Sizing Using Predictive Customer Metadata
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Sizing hardware resources for storage arrays is challenging as over-sizing leads to waste, while under-sizing results in performance bottlenecks or failure, and existing methods rely on simplistic test workloads that do not accurately reflect real-world usage.
Innovation Solution
A systematic approach using customer metadata to predict hardware needs, including SSD cache and CPU sizing, by analyzing operational parameters and application behavior, providing a reliable and accurate method to determine resource requirements through iterative processes and statistical modeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If hardware resources are oversized to ensure sufficient capacity, then reliability is improved, but loss of substance increases due to waste of funds and unused resources
Solution Approach 1:
The patent performs preliminary sizing analysis using predictive models and customer metadata before hardware deployment. The system analyzes operational parameters, application behavior, and workload characteristics to determine optimal hardware configuration in advance, preventing both over-provisioning and under-provisioning before resources are committed
Solution Approach 2:
The patent implements feedback mechanisms by continuously monitoring actual hardware utilization and comparing it against predicted values. The system uses customer metadata from operational storage arrays to refine predictive models, creating a closed-loop system that learns from real-world performance data to improve future sizing accuracy
2Loss of substance
If hardware resources are undersized to minimize waste, then loss of substance decreases, but reliability deteriorates due to performance bottlenecks or failure
Solution Approach 1:
The system performs comprehensive preliminary analysis of workload characteristics, application requirements, and operational parameters before hardware deployment. Predictive models calculate optimal resource allocation in advance, ensuring sufficient capacity is provisioned to meet performance requirements without excessive over-provisioning
Solution Approach 2:
The patent changes the approach from static, rule-based sizing to dynamic, data-driven parameter optimization. The system adjusts hardware sizing parameters based on multiple variables including application behavior patterns, workload intensity, I/O characteristics, and historical performance data to achieve optimal configuration
3Device complexity
If simplistic test workloads are used for sizing, then device complexity decreases, but measurement precision deteriorates because they do not accurately reflect real-world usage
Solution Approach 1:
The patent introduces customer metadata as an intermediary layer between test workloads and hardware sizing decisions. Instead of relying directly on simplistic test results, the system uses real-world operational data from similar customer deployments to calibrate and validate sizing predictions, bridging the gap between controlled testing and real-world performance
Solution Approach 2:
The system creates copies of real-world workload patterns and operational characteristics from customer metadata to simulate actual usage conditions during sizing analysis. Rather than using abstract test workloads, the system replicates genuine application behavior, I/O patterns, and performance characteristics observed in production environments
Data Source
AI summary
Methods and systems for enabling sizing of storage array resources are provided. Resources of a storage array can include, for example, cache, memory, SSD cache, central processing unit (CPU), storage capacity, number of hard disk drives (HDD), etc. Generally, methods and systems are provided that enable efficient predictability of sizing needs for said storage resources using historical storage array use and configuration metadata, which is gathered over time from an install base of storage arrays. This metadata is processed to produce models that are used to predict resource sizing needs to be implemented in storage arrays with certainty that takes into account customer-to-customer needs and variability. The efficiency in which the sizing assessment is made further provides significant value because it enables streamlining and acceleration of the provisioning process for storage arrays.