GPU CLUSTER · CLOUD SCHEDULING · OPTIMIZATION

GPU Cluster & Cloud Workload Optimizer

Multi-dimensional bin packing for GPU/CPU/RAM/storage/bandwidth contention. Gang scheduling, data locality, backfilling, DRF fairness, and CP-SAT placement.

Dataset Scheduler Config Gradio Bundle

Algorithms

Policies

—

Avg GPU Util %

—

Avg Queue (m)

—

Avg Fairness

Competing Scheduling Policies

These two objectives are not always aligned. Max GPU utilization may increase queue times; fairness-first policies may leave GPUs idle.

Maximize GPU Utilization

First Fit Decreasing bin packing
Best Fit fragmentation minimization
Backfilling preemptible workloads
CP-SAT gang scheduling

Minimize Wait Time & Fairness

Dominant Resource Fairness (DRF)
Min-cost flow with data locality
Priority + deadline SLA constraints
Rolling horizon online scheduling

Scenario Explorer

Pre-computed scheduling results. Select a cluster scenario.

Scenario:

Benchmark Summary

Scenario	Stress	Policy	GPU Util %	Queue (m)	Deadline Miss %	Fairness	Solve (s)

Stack

OR-Tools CP-SAT · SimPy · DRF · Min-Cost Flow · Backfilling · Rolling Horizon · Polars · Plotly · Gradio console (bundled)