GPU Usage Management via Virtual Machine Billing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In datacenter environments, users are often billed for both graphics processing and compute processing resources of GPUs, even if they only utilize one or the other, leading to potential overbilling or underutilization of GPU functionality.

Innovation Solution

A graphics processing management system that includes a processing system with a memory and a graphics processing usage management engine, which monitors GPU usage via a virtual machine system, determines the usage class using a graphics processing usage class equation, and performs actions such as usage billing or workload migration based on the identified class.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If users are billed for both graphics processing and compute processing resources of GPUs, then datacenter operators can ensure revenue coverage, but users may be overbilled for functionality they are not using

Engineering Contradiction:

Improvebilling accuracyVSAvoidoverbilling users

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The system continuously monitors GPU usage through a graphics processing usage management engine that tracks both graphics operations and compute operations in real-time. This feedback mechanism provides accurate usage data that is used to adjust billing dynamically, ensuring users are charged only for the functionality they actually utilize rather than being billed for all allocated resources

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system automatically classifies usage into different graphics processing usage classes (such as graphics-intensive, compute-intensive, or balanced) and applies appropriate billing rates without requiring manual intervention. The usage management engine self-adjusts billing based on monitored usage patterns, eliminating the need for users to manually report or justify their usage

Inventive Principle:

Principle #25Self-service

2Adaptability or versatility

If datacenter operators provision GPU functionality for both graphics and compute operations, then users have access to versatile resources, but users may not pay for functionality they are not using

Engineering Contradiction:

ImproveGPU functionality availabilityVSAvoidunderutilization of resources

Core Design Contradiction:

Adaptability or versatilityVSLoss of energy

Solution Approach 1:

The system dynamically adjusts resource provisioning and billing based on actual usage patterns. The usage management engine continuously monitors what functionality is being used and adjusts the provisioned resources accordingly, allowing the system to be versatile when needed while avoiding waste when certain functionalities are not required

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes billing parameters based on the identified graphics processing usage class. Different billing rates are applied depending on whether the usage is graphics-intensive, compute-intensive, or balanced, allowing the system to accommodate versatile GPU functionality while ensuring users pay appropriately for the specific type of operations they perform

Inventive Principle:

Principle #35Parameter changes

3Ease of operation

If conventional billing systems are used without usage monitoring, then billing processing is simple, but billing accuracy deteriorates

Engineering Contradiction:

Improvebilling process simplicityVSAvoidusage measurement accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The graphics processing usage management engine acts as an intermediary between the complex GPU usage monitoring system and the billing system. It automatically classifies usage into predefined graphics processing usage classes and determines appropriate billing rates, maintaining billing process simplicity while achieving high measurement precision through automated usage classification and monitoring

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11915060B2Graphics processing management system

Publication Date: 2024.02.27 DELL PROD LP

AI summary

A graphics processing management system includes a virtual machine coupled to client devices and a graphics processing system including graphics processing device(s). A graphics processing management subsystem is coupled to the virtual machine system and the client devices, and operates to monitor a usage of the graphics processing device(s) via the virtual machine system by a first client device. The graphics processing management subsystem then uses a graphics processing usage class equation to determine that feature(s) of the usage of the graphics processing device(s) by the first client device identifies a first graphics processing usage class. Based on determining that the usage of the graphics processing device(s) by the first client device identifies the first graphics processing usage class, the graphics processing management subsystem then performs a usage action with the first user that is associated with the first client device.