Accelerator Scheduling for Idle Resource Reallocation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Managing accelerator resources in computing environments is challenging due to inefficient utilization when workloads monopolize idle resources, leading to underutilization and pending workloads waiting for availability.

Innovation Solution

Implementing a scheduler that monitors accelerator usage and reallocates idle resources to pending workloads based on idleness conditions and priority, using a reclamation process that includes configurable thresholds and toleration times.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If accelerator resources are allocated to computing workloads, then workloads can execute tasks, but idle workloads monopolize resources causing underutilization

Engineering Contradiction:

Improveaccelerator resource utilizationVSAvoididle time of accelerators

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The system continuously monitors accelerator usage by collecting metrics from accelerators and workloads, comparing actual usage against thresholds to dynamically determine idleness and trigger reallocation when conditions are met

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The accelerator allocation state is made dynamic through continuous monitoring and evaluation, allowing the system to transition between allocated and deallocated states based on real-time workload behavior and idleness conditions

Inventive Principle:

Principle #15Dynamics

2Reliability

If workloads are assigned exclusive accelerator use, then workload execution is guaranteed, but pending workloads must wait for availability

Engineering Contradiction:

Improveworkload execution guaranteeVSAvoidwaiting time for pending workloads

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system maintains a queue of pending workloads and proactively identifies opportunities to reallocate idle accelerators before workloads are explicitly requested, enabling faster allocation to waiting workloads

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system changes the allocation parameter from exclusive guaranteed access to conditional shared access, allowing reallocation based on monitored usage patterns and idleness thresholds

Inventive Principle:

Principle #35Parameter changes

3Productivity

If a scheduler monitors and manages accelerator allocation, then resource optimization is achieved, but system complexity increases

Engineering Contradiction:

Improveresource allocation efficiencyVSAvoidscheduler complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The scheduler integrates multiple functions into a single component: monitoring accelerator usage, collecting metrics, evaluating idleness conditions, making reallocation decisions, and managing pending workloads, thereby consolidating complexity rather than distributing it

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20260064476A1Managing accelerator resources of a computer system

Publication Date: 2026.03.05 HEWLETT PACKARD ENTERPRISE DEV LP

AI summary

In certain implementations, computer-implemented method includes monitoring use of a first accelerator resource allocated to a first computing workload and determining, based on monitoring the use of the first accelerator resource allocated to the first computing workload, that the use of the first accelerator resource allocated to the first computing workload satisfies an idleness condition. The method further includes reallocating, based at least on determining that the use of the first accelerator resource allocated to the first computing workload satisfies the idleness condition, the first accelerator resource to a second computing workload, the second computing workload being a pending computing workload.