Multi-Tenant Storage Workload Homing Through Parameter Binning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Homing workloads in a multi-tenant storage environment is challenging due to the variability in user needs and resource requirements, making it difficult to identify suitable physical and virtualized resources for new or migrating workloads.

Innovation Solution

A method involving supervised learning models, specifically using a generative adversarial network, to predict suitable multi-tenant storage arrays by defining workload parameters, grouping them into bins, and training a model with historical data and test cases to infer storage array attributes, considering scaling factors for anticipated growth.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional workload homing methods are used in multi-tenant environments, then resource allocation can be achieved, but the variability in user needs and resource requirements makes it difficult to identify suitable physical and virtualized resources

Engineering Contradiction:

Improveadaptability to diverse workload requirementsVSAvoidcomplexity of resource identification process

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the workload parameters into discrete groups and bins (e.g., I/O size bins, percent read groups) to create a structured framework for classification. This segmentation transforms the continuous variability of workload requirements into manageable discrete categories that can be systematically matched with storage array characteristics.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent changes the parameters from continuous workload requirements to discrete binned categories. By transforming workload parameters into standardized bins and training the machine learning model on these binned parameters, the system achieves better adaptability to diverse workloads while reducing the complexity of the matching process.

Inventive Principle:

Principle #35Parameter changes

2Productivity

If manual resource identification processes are used, then resource allocation can be performed, but the process becomes time-consuming and inefficient for new and migrating workloads

Engineering Contradiction:

Improveworkload homing efficiencyVSAvoidtime for identifying suitable resources

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-processing workload parameters into binned categories and pre-training the machine learning model on historical workload data before actual homing decisions are needed. This preliminary preparation enables rapid homing decisions without time-consuming manual analysis when new workloads arrive or migrations are required.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces manual mechanical resource identification processes with an automated machine learning-based system. The supervised learning model automatically performs the homing decision-making, substituting human manual analysis with an automated computational system that operates faster and more consistently.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If generic storage arrays are allocated to workloads, then resource allocation is simplified, but the specific needs of diverse workloads cannot be met

Engineering Contradiction:

Improveworkload performance satisfactionVSAvoidcomplexity of matching workloads to storage arrays

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent incorporates feedback mechanisms by training the machine learning model on historical workload performance data and outcomes. The model learns from past homing decisions and their results, continuously improving its ability to match workload requirements with appropriate storage array characteristics, thereby enhancing reliability without manually increasing matching complexity.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system performs self-service by automatically analyzing workload parameters, applying the trained machine learning model, and making homing decisions without requiring manual intervention. The workload characteristics themselves drive the matching process, with the system serving itself by autonomously identifying suitable storage arrays based on the encoded workload requirements.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS12443464B2Autonomous workload homing in a multi-tenant environment

Publication Date: 2025.10.14 DELL PROD LP

AI summary

A methods for identifying a multi-tenant storage array for an application workload includes identifying workload parameters and defining a plurality of groups for each parameter and a plurality of “bins” corresponding to tuples of the groups. Exemplary workload parameters include a percent read parameter and an I/O size parameter. A bin mix of the workload is determined based on historical data wherein the bin mix indicates bins associated with workload activity exceeding a specified threshold. The bin mix is used to define at least some inputs for a supervised learning model of a process for homing application workloads in a multi-tenant storage array. After appropriate training of the model with a generative adversarial network, the model may be invoked to infer or predict attributes of a suitable storage array. The workload may be associated with a scaling factor that influences the determination of a suitable storage array.