The paper presents a fine‑grained monitoring study of a university GPU cluster, collecting data at 100 ms intervals using a lightweight tool built on Nvidia’s DCGM library. Analysis of a 10‑day period with 7,399 jobs shows that over 70 % of jobs use less than 20 % of GPU memory and 50 % have below 5 % SM utilization, while 80 % of jobs experience a large mismatch between requested and actual runtimes. The authors also observe highly periodic communication patterns in multi‑GPU jobs, indicating that significant performance gains could be achieved through improved scheduling and multitasking rather than hardware upgrades.
GPU clusters have become essential computing resources for AI and scientific research in academia and industry. However, their high cost and growing demand often result in long queueing delays. While...
The paper introduces PWR, a scheduling policy that selects power‑efficient GPU and CPU combinations to reduce power consumption in GPU datacenters. It complements the existing Fragmentation Gradient Descent (FGD) policy, which mitigates GPU fragmentation by using a fragmentation metric. Experiments in a simulated cluster show that combining PWR with FGD balances power savings and fragmentation reduction.
arXiv:2605. 09370v3 Announce Type: replace-cross Abstract: Large-scale AI training is now fundamentally a distributed systems problem, and hardware failures have become routine operating conditions rather than rare exceptions.
arXiv:2610.28223v1 Announce Type: cross Abstract: Large language models can solve complex problems across a variety of subjects, such as mathematics and coding. For such complex analytical tasks, mod...
arXiv:2508.19073v4 Announce Type: replace-cross Abstract: Deep learning training commonly runs on shared multi-tenant GPU servers, where exclusive allocation provides isolation but can leave resource...
arXiv:2604. 10180v2 Announce Type: replace-cross Abstract: Disaggregation maps parts of an AI workload to different types of GPUs, offering a path to utilize modern heterogeneous GPU clusters.
arXiv:2606. 00735v1 Announce Type: cross Abstract: In distributed Mixture-of-Experts (MoE) inference, input-dependent token routing interacts with GPU performance variability to create persistent stragglers under synchronized execution, where the slowest GPU determines layer latency.
arXiv:2606. 09200v1 Announce Type: cross Abstract: The rapid growth of large-scale machine learning (ML) has made distributed training across multiple GPUs a fundamental component of modern ML systems.
arXiv:2610.14009v1 Announce Type: cross Abstract: Large AI models increasingly rely on distributed GPU execution, yet it remains unclear whether established failure analysis methods transfer effectiv...
The paper evaluates a learned request‑routing policy for disaggregated large‑language‑model serving, where compute‑heavy prefill and memory‑heavy decode stages run on separate GPU pools. Using a discrete‑event simulator and real NVIDIA A40 GPUs, the calibrated router—leveraging prompt length, predicted output length, KV‑cache pressure, and SLO class—outperforms round‑robin, least‑loaded, and length‑based heuristics, achieving the highest mean goodput (0.864) and lowest variance across three mixed, bursty arrival traces. Hardware calibration proves critical, providing a 4.5‑point goodput boost and roughly 40 % of the tail‑latency advantage, and the learned router can match round‑robin performance with one fewer GPU in certain scenarios.
The paper introduces REACT, a system that dynamically tunes communication collectives in distributed AI training to mitigate congestion without requiring network infrastructure changes. REACT operates at the application layer, detecting congestion via flow statistics and adjusting the pattern of data exchange—such as selecting different aggregation nodes in an AllReduce tree—while preserving the semantics of the communication. Evaluations on a shared academic GPU cluster show that REACT improves algorithm bandwidth by 13%–38% under congestion, with simulations indicating potential gains up to 75%.