Deploying heterogeneous AI models concurrently on a shared GPU introduces resource contention that complicates runtime scheduling. While surrogate models avoid costly online benchmarking, their profiling requirements typically grow combinatorially with the number of co-running models, limiting scalability. We propose a MeanField surrogate that predicts per-model performance from local configuration and aggregate GPU state rather than explicitly modeling all joint interactions. Experiments on concurrent LLM and vision workloads across $N \in \{2,3,4,5,6\}$ show high predictive accuracy ($R^2 \approx 0.96$) with an empirical sample budget that grows approximately linearly in $N$, in contrast to the combinatorial cost of fully joint profiling. Integrated into a genetic algorithm scheduler, the surrogate scales to an $N=5$ problem with 78,732 feasible joint configurations, remaining within 0.10% of the exhaustive search with zero SLA violations across eight dynamic workload scenarios, while complete online GA decisions take 26 ms median, about $5\times$ faster than exhaustive surrogate search.
Container-granularity scheduling leaves abundant short-lived idle slices within containers unexploited. Reallocating containers is too heavyweight to utilize such fine-grained opportunities under SLA constraints, and operator-level scheduling requires reasoning about dependencies, memory safety, and cluster-wide execution dynamics in real time. In this paper, we present SliceScheduler, a dynamic operator-level scheduling system for multi-tenant model serving. The key idea is to expose cluster-wide operator execution state and enable what-if reasoning over scheduling decisions. SliceScheduler consists of four key components. First, we introduce the Global Mapping Graph (GMG), a unified abstraction that captures operator dependencies, tensor shapes, resource mappings, and execution states, providing a real-time, cluster-wide view with explicit resource semantics. Second, we build a global simulator on top of GMG to predict operator-level execution and memory evolution under candidate placements. Third, we design an incremental, simulation-based scheduling module that selects placements to exploit fragmented idle slices while avoiding memory violations and preserving SLA. Finally, we develop an operator executor that materializes scheduling decisions on GPUs and coordinates computation and cross-accelerator transfers. We implement SliceScheduler as a PyTorch backend and evaluate it using production trace replay. Experimental results show that SliceScheduler improves token throughput by 1.10--2.29$\times$ compared to existing approaches, while maintaining SLA violations within 9\%. SliceScheduler demonstrates that operator-level scheduling is a practical and effective approach to improving GPU utilization for multi-tenant LLM serving.
Vision-language models (VLMs) enable embodied agents to reason and act from visual observations and language instructions. Reinforcement learning (RL) post-training enhances these capabilities using task feedback, but current on-policy RL runtimes execute rollout, reference scoring, and actor training in strict serial phases. While effective for text-only RL, this phase-granular execution is wasteful for VLMs, where processing dense video inputs and prompt prefixes occupies a large fraction of each phase. Because prefix processing is independent of the generated response, it can be run alongside rollout decoding, which leaves GPU compute capacity underutilized, without breaking synchronous on-policy semantics. We present Rollplex, a runtime that decomposes the reference and training phase and moves the prefix computation into the rollout decode window. Realizing this schedule requires more than concurrent kernel launches: naive colocation of Qwen2.5-VL-32\,B requires roughly 165\,GiB per GPU, while rollout and training prefer different tensor-parallel (TP) degrees and weight layouts. Rollplex addresses these constraints with two mechanisms. Phase-aware memory management controls HBM residency according to producer--consumer lifetimes. Parallelism-aware weight sharing uses the same physical storage for layout-compatible tensors across distinct TP degrees and reconstructs only incompatible tensors, avoiding a complete second actor copy. On 32 H800 GPUs, Rollplex achieves $1.23\times$--$1.30\times$ speedup over serial colocation and $1.57\times$--$2.24\times$ over disaggregation under the same GPU budget, while preserving the synchronous RL update.
Text-to-image (T2I) workflows are increasingly deployed on serverless platforms because users often compose customized workflows and invoke them intermittently. Existing platforms typically deploy each workflow as an opaque GPU function, provisioning, placing, and scaling all constituent models in the workflow together. This monolithic design obscures workflow structure, inflates scaling overhead, forces users to manage low-level GPU coordination, and limits fine-grained fairness in multi-tenant clusters. In this paper, we present ServerlessT2I, a serverless-native system that decomposes a T2I workflow into loosely coupled model functions that can be independently managed and scheduled. By explicitly managing individual model execution, ServerlessT2I enables per-model scaling, declarative workflow composition, transparent GPU-resident communication, and fairness-aware scheduling. To make this decomposition efficient, ServerlessT2I harvests slack GPU memory left idle by compute-bound T2I inference to build a data plane that reduces model loading and data communication overheads. \sys{} further introduces a fair scheduler for multi-tenant serving. Using production traces, ServerlessT2I sustains up to 2$\times$ higher request rates than existing T2I workflow serving systems with the same GPU budget; for a fixed request rate, it saves up to 3$\times$ GPU resources while satisfying service level objectives (SLOs).
Minyu Cui, Anna Wingkvist, Morgan Ericssoncs.DC cs.AI
Mixture-of-Experts (MoE) architectures increase model capacity without proportionally increasing computation cost and have become a key building block for scaling large language models (LLMs) to trillion-parameter regimes. Efficient deployment of these MoE models relies on distributed execution across multiple GPUs, where each MoE layer involves two all-to-all communications: dispatching tokens to expert ranks and returning the expert outputs to their source ranks. Conventional MoE implementations launch this return all-to-all after expert compute completes, exposing communication latency on the critical path and reducing GPU utilization. We present a fine-grained approach that overlaps expert compute with the second all-to-all via tile-level signaling and scheduling. Our producer-consumer co-design combines: (1) a persistent per-rank computation kernel (producer) that covers all local experts on the rank to eliminate repeated kernel launch overhead and prioritizes remote-critical tiles, and (2) a persistent communication kernel (consumer) on a small dedicated partition of streaming multiprocessors (SMs) that issues segment-granular transfers as tiles become ready. Our co-design avoids intrusive changes to the underlying computation operators or communication primitives, making it practical for improving distributed MoE execution efficiency on multi-GPU systems. On a 4-A100 GPU platform, evaluated on three MoE models against four state-of-the-art MoE systems, our approach achieves up to 2.64x end-to-end speedup and 2.74x MoE-layer speedup. Compared with a conventional non-overlap baseline, our approach consistently improves both operator- and MoE-layer-level performance across varying GEMM shapes, router modes, and a broad range of producer/consumer SM partitions, while preserving correctness.
Ashiyana Abdul Majeed, Mahmoud Meribout, Neethu Joseph +2cs.AR cs.AI
Physical AI systems, such as autonomous vehicles and intelligent machines, require transformer-based perception models that satisfy stringent edge latency and energy constraints. However, heterogeneous edge-GPU deployment remains limited by underutilized hardware engines and accelerator-incompatible operators, causing fragmented execution and lower throughput per watt. This paper presents Heterogeneous Frame Dispatch Scheduling (H-FraDS), a hardware-aware frame scheduling methodology for transformer inference on a recent NVIDIA edge GPU. H-FraDS routes frames across the GPU and dual deep learning accelerator (DLA) cores using fixed dispatch ratios to improve utilization under latency and power constraints. To enable scheduling, incompatible transformer components are adapted for DLA execution by reshaping tensors, approximating error function (ERF) with tanh, and replacing layer normalization with bounded tanh. The adapted model maintains a 92% F1 score, with only a 2% reduction from the original. Optical flow accelerator (OFA) is further used for inference-side optical-flow estimation. To the best of the authors' knowledge, prior work has not addressed these combined issues. Using Swin Transformer for autonomous-driving perception, H-FraDS Balanced Dispatch (1:2) achieves 125.93 FPS, a 2.36x speedup over standalone adapted-DLA execution, 4.0 FPS/W, and approximately 24 ms DLA latency, satisfying 30 FPS real-time operation; the GPU-DLA-OFA case achieves a 2.02x DLA throughput speedup.
A persistent interactive world model keeps its running state resident on the GPU that serves it: a multi-gigabyte attention cache, almost all of it rewritten at every generation step. That state cannot be recomputed in interactive time or approximated without changing the world, so a live session pins its device. The pin is a scheduling problem. WorldMove moves a live session under one guarantee: the destination is bit-identical to the source, or nothing is installed. It relocates the cache in 18.8 ms same-node, 101x faster than save/load. It holds a checksum-verified 92.1-94.8 Gb/s on a 100 Gb fabric. At that rate the cache fits inside one interactive block. Migrating an actively generating session, it converges at a block boundary and the destination continues the world bit for bit. An admissibility condition decides each move. The move must complete inside the readout horizon, over bandwidth that covers the state plus its dirty rate. Lifted to a fleet schedulability test, it governed a consolidation loop that executed 48 of 48 migrations bit-identical across two providers. Two constraints are structural. Bit-exactness survives only inside a controlled configuration of one GPU architecture, so moving the state is the only way to preserve it exactly in interactive time. Verification cannot hide inside the wire on this fabric. Receive-path checksums stall the transport at protocol timescales under fan-in, and unscheduled incast silently collapses a receiver while every delivered byte stays correct. An incast-aware admission controller holds zero misses to 1.4x offered load and sheds overload as rejects. A lossless GPU codec widens the admission gate to fabrics raw motion cannot use. We exercise the serving loop and the mover separately, each end to end. Their composition on one fabric is unbuilt. Exact-state elasticity is a joint scheduling problem over transport and verification.
Balasubramanian Sivan, Renato Paes Leme, Mihai Tiuca +6cs.LG cs.GT
The escalating demand for Machine Learning (ML) training resources in recent years has resulted in a substantial gap between the high demand and the available supply. Efficient allocation of these scarce and expensive resources is crucial for organizations to maximize their return on investment. Existing resource allocation mechanisms, like Karma [OSDI'23], are designed to guarantee Pareto efficiency and max-min fairness in settings with dynamic (time-varying) user demands, but fail to preserve these key properties in the presence of demands with heterogeneous values. Given the ubiquity and inevitability of heterogeneity in organizational values of different workloads, effective resource allocation policies must accommodate these variations. In this paper, we describe the design, implementation, deployment, and theoretical analysis of Quota Marketplace, a market-based mechanism to efficiently allocate ML training chips (like GPUs), explicitly addressing scenarios with demands of heterogeneous value. We detail the implementation of this mechanism within Google and present metrics that demonstrate its impact. We also discuss many business-critical requirements that the Quota Marketplace handles quite effectively, and document the gains and opportunities it has unlocked. We establish theoretically how this market-based approach achieves the essential properties of Pareto efficiency and max-min fairness by allowing the users to express the value of their workloads and enabling dynamic resource pricing based on supply and demand fluctuations. Ultimately, the market facilitates resource allocation that aligns with organizational priorities.
RL-based LLM post-training increasingly disaggregates Rollout and Training across separate GPU resources, but static GPU partitioning suffers from severe pipeline bubbles under long-tail rollout latency. We present DynaResize, a runtime GPU reallocation system that dynamically switches GPUs between Rollout and Training to balance stage execution times without changing RL semantics. DynaResize decomposes resizing into fine-grained operations and removes non-startup-critical work from the critical path through communicator reuse, bounded state staging, and hysteresis-based resizing. Experimental results show that DynaResize can improve end-to-end throughput by 66.5% and reduce total execution time by 33% over the optimal static configuration, while hiding 27% of role-switching overhead.
Kaiwen Chen, Xin Tan, Jingzong Li +1cs.LG cs.AI cs.DC
Reinforcement learning (RL) has emerged as a standard post-training paradigm for shaping large language models (LLMs) into capable agents. In agentic RL, the rollout stage generates trajectories while invoking tools, producing long-tailed and non-stationary workloads that expose two fundamental challenges in resource management. First, due to the long-tail distribution, a small fraction of trajectories dominates rollout makespan. Second, rollout and training are subject to cross-stage imbalance, as they exhibit strong asymmetry in compute patterns, memory demands, and sensitivity to sequence length. Compounding this asymmetry, the sequence length distribution drifts continuously as the policy evolves, rendering any static resource split progressively suboptimal. We present Libra, a resource management system to address both challenges via two core mechanisms. The first is a global resource planner that jointly optimizes GPU allocation across rollout and training clusters. It leverages an elastic hybrid pool to enable lightweight, non-blocking worker reallocation between stages. The second is a causality-driven multi-level feedback queue (C-MLFQ) scheduler, which routes requests to heterogeneous rollout buckets based on causal signals derived from tool-return outcomes, rather than relying on fragile length predictions. Evaluated on 48 A800 GPUs, Libra achieves up to 3.0x higher throughput and converges up to 2.5x faster in reward compared to the baselines.
The rapid growth of large language model (LLM) inference services has increased the demand for efficient multi-tenant GPU scheduling. While modern inference runtimes such as vLLM improve throughput through continuous batching and optimized memory management, accurately estimating the runtime cost of heterogeneous inference requests remains challenging. In practice, admission-time workload estimates may deviate from observed execution behavior, leading to workload misclassification, queue imbalance, increased tail latency, and degraded Quality-of-Service (QoS). This paper presents DriftSched, a QoS-aware scheduling framework for multi-tenant LLM inference serving on NVIDIA L4 GPUs. DriftSched combines workload classification, token-budget estimation, tenant-aware queue management, and an online feedback mechanism to refine workload estimates using runtime observations. The framework evaluates FIFO, Priority, Weighted, Shortest-Job-First (SJF), and Aging Priority scheduling policies under heterogeneous multi-tenant workloads. Experimental results show that adaptive calibration reduces workload estimation error by an average of 38.8% (MAE) and 40.5% (RMSE), improving workload classification stability. Among all evaluated schedulers, SJF achieves the best overall performance, reducing median end-to-end latency by approximately 42% and P99 latency by approximately 16% relative to FIFO under sustained GPU contention. The results further indicate that scheduler selection has a greater impact on latency behavior than runtime calibration alone, while accurate workload characterization largely eliminates systematic estimation drift. This work contributes a reproducible framework for studying workload-estimation fidelity and QoS-aware scheduling in multi-tenant GPU inference systems.
Machine learning (ML) inference serving systems host deep neural network (DNN) models and schedule incoming inference requests across deployed GPUs. However, limited support for task prioritization and insufficient latency estimation under concurrent execution may restrict their applicability in on-premises scenarios. We present \emph{Strait}, a serving system designed to enhance deadline satisfaction for dual-priority inference traffic under high GPU utilization. To improve latency estimation, Strait models potential contention during data transfer and accounts for kernel execution interference through an adaptive prediction model. By drawing on these predictions, it performs priority-aware scheduling to deliver differentiated handling. Evaluation results under intense workloads suggest that Strait reduces deadline violations for high-priority tasks by 1.02 to 11.18 percentage points while incurring acceptable costs on low-priority tasks. Compared to software-defined preemption approaches, Strait also exhibits more equitable performance.