Compute Instance Scheduling via QoS Feedback and Interference Coefficients

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing compute instance scheduling methods in public cloud scenarios lack precision, leading to suboptimal utilization of physical resources and quality of service (QoS) for tenants, as they fail to accurately manage resource contention and varying service requirements across multiple compute instances on shared hardware.

Innovation Solution

The proposed method introduces the collection and calculation of QoS parameters for both compute instances and physical resources, allowing for precise scheduling by identifying and controlling compute instances exceeding QoS thresholds, and determining interference coefficients to optimize resource allocation and migration strategies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If existing compute instance scheduling methods are used, then the system can provide compute instances for tenants, but the scheduling precision is low leading to suboptimal physical resource utilization

Engineering Contradiction:
Improvescheduling precisionVSAvoidphysical resource utilization
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent implements a feedback mechanism by calculating QoS parameters for both compute instances and physical resources, using these parameters to determine interference coefficients, and adjusting scheduling decisions based on this feedback to improve both precision and resource utilization

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent changes the scheduling parameters from simple resource allocation to a comprehensive system that incorporates QoS parameters, interference coefficients, and compute instance characteristics, enabling precise scheduling that optimizes physical resource utilization

Inventive Principle:
Principle #35Parameter changes

2Productivity

If more compute instances are scheduled on same hardware, then physical resource utilization increases, but QoS of the service run by the tenant deteriorates

Engineering Contradiction:
Improvephysical resource utilizationVSAvoidquality of service
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent applies local quality by calculating and considering individual QoS parameters for each computeinstance and physical resource, allowing differentiated scheduling decisions that ensure high QoS for critical tenants while maximizing overall resource utilization

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system uses QoS parameter feedback to continuously monitor service quality and adjust scheduling decisions, ensuring that increasing resource utilization does not compromise tenant service quality

Inventive Principle:
Principle #23Feedback

3Ease of operation

If compute instances are scheduled without considering physical resource usage, then scheduling is simple, but resource contention and service requirements are not accurately managed

Engineering Contradiction:
Improvescheduling simplicityVSAvoidresource contention management precision
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent introduces feedback loops that continuously collect physical resource usage data and QoS parameters, processing this information to determine interference coefficients and adjust scheduling decisions for accurate resource contention management

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The system transforms scheduling from a simple process to a parameter-driven approach that incorporates QoS metrics, interference coefficients, and physical resource states, enabling precise management of resource contention while maintaining operational automation

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3825854B1Compute instance scheduling method and apparatus
Publication Date: 2025.01.08 HUAWEI CLOUD COMPUTING TECHNOLOGIES CO LTD
  • EP3825854B1 patent drawingFigure 1
  • EP3825854B1 patent drawingFigure 2
  • EP3825854B1 patent drawingFigure 3

AI summary

This application discloses a compute instance scheduling method, and an apparatus and a device for performing the method. According to the method, usage of a physical resource included in each computer, and usage of a physical resource occupied by a compute instance running on each computer in a compute instance provisioning system are collected. Then, a compute instance in the compute instance provisioning system is scheduled based on the usage of the physical resource in each computer and the usage of the physical resource occupied by the compute instance running on each computer. Therefore, the method improves precision of scheduling the compute instance and utilization of a resource in a computer.