An official website of the United States government
Real-time applications such as autonomous and connected cars, surveillance, and online learning applications have to train on streaming data. They require low-latency, high throughput machine learning (ML) functions resident in the network and in the cloud to perform learning and inference. NFV on edge cloud platforms can provide support for these applications by having heterogeneous computing including GPUs and other accelerators to offload ML-related computation. GPUs provide the necessary speedup for performing learning and inference to meet the needs of these latency sensitive real-time applications. Supporting ML inference and learning efficiently for streaming data in NFV platforms has several challenges. In this paper, we present a framework, NetML, that runs existing ML applications on an heterogeneous NFV platform that includes both CPUs and GPUs. NetML efficiently transfers the appropriate packet payload to the GPU, minimizing overheads, avoiding locks, and avoiding CPU-based data copies. Additionally, NetML minimizes latency by maximizing overlap between the data movement and GPU computation. We evaluate the efficiency of our approach for training and inference using popular object detection algorithms on our platform. NetML reduces the latency for inferring images by more than 20% and increases the training throughput by 30% while reducing CPU utilization compared to other state-of-the-art alternatives.
more »
« less
This content will become publicly available on July 9, 2026
PPipe: Efficient Video Analytics Serving on Heterogeneous GPU Clusters via Pool-Based Pipeline Parallelism
With the rapid innovation of GPUs, heterogeneous GPU clusters in both public clouds and on-premise data centers have become increasingly commonplace. In this paper, we demonstrate how pipeline parallelism, a technique wellstudied for throughput-oriented deep learning model training, can be used effectively for serving latency-bound model inference, e.g., in video analytics systems, on heterogeneous GPU clusters. Our work exploits the synergy between diversity in model layers and diversity in GPU architectures, which results in comparable inference latency for many layers when running on low-class and high-class GPUs. We explore how such overlooked capability of low-class GPUs can be exploited using pipeline parallelism and present a novel inference serving system, PPipe, that employs pool-based pipeline parallelism via an MILP-based control plane and a data plane that performs resource reservation-based adaptive batching. Evaluation results on diverse workloads (18 CNN models) show that PPipe achieves 41.1%–65.5% higher utilization of low-class GPUs while maintaining high utilization of high-class GPUs, leading to 32.2%–75.1% higher serving throughput compared to various baselines.
more »
« less
- PAR ID:
- 10643394
- Publisher / Repository:
- USENIX
- Date Published:
- Subject(s) / Keyword(s):
- Heterogeneous GPU Clusters, Pool-based pipeline parallelism, video analytics serving
- Format(s):
- Medium: X
- Location:
- Boston, MA
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Today's data centers often need to run various machine learning (ML) applications with stringent SLO (Service-Level Objective) requirements, such as inference latency. To that end, data centers prefer to 1) over-provision the number of servers used for inference processing and 2) isolate them from other servers that run ML training, despite both use GPUs extensively, to minimize possible competition of computing resources. Those practices result in a low GPU utilization and thus a high capital expense. Hence, if training and inference jobs can be safely co-located on the same GPUs with explicit SLO guarantees, data centers could flexibly run fewer training jobs when an inference burst arrives and run more afterwards to increase GPU utilization, reducing their capital expenses. In this paper, we propose GPUColo, a two-tier co-location solution that provides explicit ML inference SLO guarantees for co-located GPUs. In the outer tier, we exploit GPU spatial sharing to dynamically adjust the percentage of active GPU threads allocated to spatially co-located inference and training processes, so that the inference latency can be guaranteed. Because spatial sharing can introduce considerable overheads and thus cannot be conducted at a fine time granularity, we design an inner tier that puts training jobs into periodic sleep, so that the inference jobs can quickly get more GPU resources for more prompt latency control. Our hardware testbed results show that GPUColo can precisely control the inference latency to the desired SLO, while maximizing the throughput of the training jobs co-located on the same GPUs. Our large-scale simulation with a 57-day real-world data center trace (6500 GPUs) also demonstrates that GPUColo enables latency-guaranteed inference and training co-location. Consequently, it allows 74.9% of GPUs to be saved for a much lower capital expense.more » « less
-
null (Ed.)Deep neural networks (DNNs) are increasingly used for real-time inference, requiring low latency, but require significant computational power as they continue to increase in complexity. Edge clouds promise to offer lower latency due to their proximity to end-users and having powerful accelerators like GPUs to provide the computation power needed for DNNs. But it is also important to ensure that the edge-cloud resources are utilized well. For this, multiplexing several DNN models through spatial sharing of the GPU can substantially improve edge-cloud resource usage. Typical GPU runtime environments have significant interactions with the CPU, to transfer data to the GPU, for CPU-GPU synchronization on inference task completions, etc. These result in overheads. We present a DNN inference framework with a set of software primitives that reduce the overhead for DNN inference, increase GPU utilization and improve performance, with lower latency and higher throughput. Our first primitive uses the GPU DMA effectively, reducing the CPU cycles spent to transfer the data to the GPU. A second primitive uses asynchronous ‘events’ for faster task completion notification. GPU runtimes typically preclude fine-grained user control on GPU resources, causing long GPU downtimes when adjusting resources. Our third primitive supports overlapping of model-loading and execution, thus allowing GPU resource re-allocation with very little GPU idle time. Our other primitives increase inference throughput by improving scheduling and processing more requests. Overall, our primitives decrease inference latency by more than 35% and increase DNN throughput by 2-3×.more » « less
-
Large language models have led to state-of-the-art accuracies across several tasks. However, training these models efficiently is challenging because: a) GPU memory capacity is limited, making it impossible to fit large models on even a multi-GPU server, and b) the number of compute operations required can result in unrealistically long training times. Consequently, new methods of model parallelism such as tensor and pipeline parallelism have been proposed. Unfortunately, naive usage of these methods leads to scaling issues at thousands of GPUs. In this paper, we show how tensor, pipeline, and data parallelism can be composed to scale to thousands of GPUs. We propose a novel interleaved pipelining schedule that can improve throughput by 10+% with memory footprint comparable to existing approaches. Our approach allows us to perform training iterations on a model with 1 trillion parameters at 502 petaFLOP/s on 3072 GPUs (per-GPU throughput of 52% of theoretical peak).more » « less
-
Neural-network-enabled data analysis in real-time scientific applications imposes stringent requirements on inference latency. Meanwhile, recent deep learning (DL) model design trends to replace a single branch with multiple branches for high prediction accuracy and robustness, which makes interoperator parallelization become an effective approach to improve inference latency. However, existing inter-operator parallelization techniques for inference acceleration are mainly focused on utilization optimization in a single GPU. With the data size of an input sample and the scale of a DL model ever-growing, the limited resource of a single GPU is insufficient to support the parallel execution of large operators. In order to break this limitation, we study hybrid inter-operator parallelism both among multiple GPUs and in each GPU. In this paper, we design and implement a hierarchical inter-operator scheduler (HIOS) to automatically distribute large operators onto different GPUs and group small operators in the same GPU for parallel execution. Particularly, we propose a novel scheduling algorithm, named HIOS-LP, which consists of inter-GPU operator parallelization through iterative longest-path (LP) mapping and intra-GPU operator parallelization based on a sliding window. In addition to extensive simulation results, experiments with modern convolutional neural network benchmarks demonstrate that our HIOS-LP outperforms the state-of-the-art inter-operator scheduling algorithm IOS by up to 17% in real systems.more » « less
