What does this research mean for the field?

The H-GEMA scheduling architecture eliminates real-time SLA violations during training bursts and reduces topological dispersion by 31-50% in heterogeneous GPU clusters compared to greedy baselines. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To develop and evaluate H-GEMA, a five-phase scheduling architecture balancing real-time inference latency, batch training throughput, topology-aware placement, and starvation prevention on heterogeneous GPU clusters.

August 30, 2026Open Access

H-GEMA: Hybrid Geo-stochastic Error-Mitigating Architecture for GPU Cluster Scheduling

View Full Paper

Ask AI

Bookmark

To develop and evaluate H-GEMA, a five-phase scheduling architecture balancing real-time inference latency, batch training throughput, topology-aware placement, and starvation prevention on heterogeneous GPU clusters.
Integrated a Lyapunov drift-plus-penalty Virtual Queue Tracker, an online A-NWIN Whittle-index neural estimator, and a Bertsekas Auction-based GPU assignment algorithm.
Implemented hardware-level constraints including topology-aware placement costs, thermal filtering, and VRAM safety validation.
Evaluated the architecture using tick-based simulation on NVIDIA A100 GPU clusters across four scenarios: training bursts, normal mixed operations, constrained capacity, and priority flooding.
Eliminated real-time SLA violations during training bursts (0% for H-GEMA vs. 11.51% for the greedy baseline) and substantially mitigated violations during extreme priority flooding.
Reduced GPU topological dispersion by approximately 31% to 50% across three workload scenarios relative to baseline scheduling.
Identified that sustained mixed operations can allow long-duration multi-GPU training jobs to monopolize cluster capacity, delaying subsequent real-time tasks.

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

Simulation study demonstrates eliminated latency violations and reduced hardware dispersion in heterogeneous GPU clusters, highlighting the benefits of stochastic multi-factor scheduling.

Cite This Study

Éric GUSTAVO REIS DE SENA (2026) studied this question.

synapsesocial.com/papers/6a93f0a36c1a8fb52e79cf99 https://doi.org/10.5281/zenodo.22144744

View Full Paper

Ask AI

Bookmark

Implication

Simulation study demonstrates eliminated latency violations and reduced hardware dispersion in heterogeneous GPU clusters, highlighting the benefits of stochastic multi-factor scheduling.

H-GEMA: Hybrid Geo-stochastic Error-Mitigating Architecture for GPU Cluster Scheduling

Key Points

Discussion

Implication

Cite This Study

H-GEMA: Hybrid Geo-stochastic Error-Mitigating Architecture for GPU Cluster Scheduling

Key Points

Discussion

Implication

Cite This Study

Also Consider

Also Consider