skip to main content
US FlagAn official website of the United States government
dot gov icon
Official websites use .gov
A .gov website belongs to an official government organization in the United States.
https lock icon
Secure .gov websites use HTTPS
A lock ( lock ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

Explore Research Products in the PAR
It may take a few hours for recently added research products to appear in PAR search results.


Title: MAPA: multi-accelerator pattern allocation policy for multi-tenant GPU servers
Multi-accelerator servers are increasingly being deployed in shared multi-tenant environments (such as in cloud data centers) in order to meet the demands of large-scale compute-intensive workloads. In addition, these accelerators are increasingly being inter-connected in complex topologies and workloads are exhibiting a wider variety of inter-accelerator communication patterns. However, existing allocation policies are ill-suited for these emerging use-cases. Specifically, this work identifies that multi-accelerator workloads are commonly fragmented leading to reduced bandwidth and increased latency for inter-accelerator communication. We propose Multi-Accelerator Pattern Allocation (MAPA), a graph pattern mining approach towards providing generalized allocation support for allocating multi-accelerator workloads on multi-accelerator servers. We demonstrate that MAPA is able to improve the execution time of multi-accelerator workloads and that MAPA is able to provide generalized benefits across various accelerator topologies. Finally, we demonstrate a speedup of 12.4% for 75th percentile of jobs with the worst case execution time reduced by up to 35% against baseline policy using MAPA.  more » « less
Award ID(s):
2047521
PAR ID:
10319195
Author(s) / Creator(s):
; ; ; ;
Date Published:
Journal Name:
SC'21: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; ; ; ; ; ; (, USENIX)
    Machine learning models are increasingly being trained across multiple GPUs and servers. In this setting, data is transferred between GPUs using communication collectives such as ALLTOALL and ALLREDUCE, which can become a significant bottleneck in training large models. Thus, it is important to use efficient algorithms for collective communication. We develop TACCL, a tool that enables algorithm designers to guide a synthesizer into automatically generating algorithms for a given hardware configuration and communication collective. TACCL uses a novel communication sketch abstraction to get crucial information from the designer to significantly reduce the search space and guide the synthesizer towards better algorithms. TACCL also uses a novel encoding of the problem that allows it to scale beyond single-node topologies. We use TACCL to synthesize algorithms for three collectives and two hardware topologies: DGX-2 and NDv2. We demonstrate that the algorithms synthesized by TACCL outperform the Nvidia Collective Communication Library (NCCL) by up to 6.7x. We also show that TACCL can speed up end-to-end training of Transformer-XL and BERT models by 11%–2.3x for different batch sizes. 
    more » « less
  2. ; ; ; ; (, ACM Transactions on Modeling and Computer Simulation)
    In high-performance computing (HPC), modern supercomputers typically provide exclusive computing resources to user applications. Nevertheless, the interconnect network is a shared resource for both inter-node communication and across-node I/O access, among co-running workloads, leading to inevitable network interference. In this study, we develop MFNetSim, a multi-fidelity modeling framework that enables simulation of multi-traffic simultaneously over the interconnect network, including inter-process communication and I/O traffic. By combining different levels of abstraction, MFNetSim can efficiently co-model the communication and I/O traffic occurring on HPC systems equipped with flash-based storage. We conduct simulation studies of hybrid workloads composed of traditional HPC applications and emerging ML applications on a 1,056-node Dragonfly system with various configurations. Our analysis provides various observations regarding how network interference affects communication and I/O traffic. 
    more » « less
  3. ; ; ; ; (, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems)
    FPGA-based edge servers are used in many applications in smart cities, hospitals, retail, etc. Equipped with heterogeneous FPGA-based accelerator cards, the servers can be implemented with multiple tasks including efficient video prepossessing, machine learning algorithm acceleration, etc. These servers are required to implement inference during the daytime while re-training the model during the night to adapt to new environments, domains, or new users. During the re-training, conventionally, the incoming data are transmitted to the cloud, and then the updated machine learning models will be transferred back to the edge server. Such a process is inefficient and cannot protect users’ privacy, so it is desirable for the models to be directly trained on the edge servers. Deploying convolutional neural network (CNN) training on heterogeneous resource-constrained FPGAs is challenging since it needs to consider both the complex data dependency of the training process and the communication bottleneck among different FPGAs. Previous multi-accelerator training algorithms select optimal scheduling strategies for data parallelism, tensor parallelism, and pipeline parallelism. However, pipeline parallelism cannot deal with batch normalization (BN) which is an essential CNN operator, while purely applying data parallelism and tensor parallelism suffers from resource under-utilization and intensive communication costs. In this work, we propose MTrain, a novel multi-accelerator training scheduling strategy that transfers the training process into a multi-branch workflow, thus independent sub-operations of different branches are executed on different training accelerators in parallelism for better utilization and reduced communication overhead. Experimental results show that we can achieve efficient CNN training on heterogeneous FPGA-based edge servers with 1.07x-2.21x speedup under 15 GB/s peer-to-peer bandwidth compared to the state-of-the-art work. 
    more » « less
  4. ; ; ; (, IEEE Circuits and System)
    Generative adversarial networks (GANs) have emerged as a powerful solution for generating synthetic data when the availability of large, labeled training datasets is limited or costly in large-scale machine learning systems. Recent advancements in GAN models have extended their applications across diverse domains, including medicine, robotics, and content synthesis. These advanced GAN models have gained recognition for their excellent accuracy by scaling the model. However, existing accelerators face scalability challenges when dealing with large-scale GAN models. As the size of GAN models increases, the demand for computation and communication resources during inference continues to grow. To address this scalability issue, this article proposes Chiplet-GAN, a chiplet-based accelerator design for GAN inference. Chiplet-GAN enables scalability by adding more chiplets to the system, thereby supporting the scaling of computation capabilities. To handle the increasing communication demand as the system and model scale, a novel interconnection network with adaptive topology and passive/active network links is developed to provide adequate communication support for Chiplet-GAN. Coupled with workload partition and allocation algorithms, Chiplet-GAN reduces execution time and energy consumption for GAN inference workloads as both model and chiplet-system scales. Evaluation results using various GAN models show the effectiveness of Chiplet-GAN. On average, compared to GANAX, SpAtten, and Simba, the Chiplet-GAN reduces execution time and energy consumption by 34% and 21%, respectively. Furthermore, as the system scales for large-scale GAN model inference, Chiplet-GAN achieves reductions in execution time of up to 63% compared to the Simba, a chiplet-based accelerator. 
    more » « less
  5. ; ; ; (, IEEE circuits and systems magazine)
    Generative adversarial networks (GANs) have emerged as a powerful solution for generating synthetic data when the availability of large, labeled training datasets is limited or costly in large-scale machine learning systems. Recent advancements in GAN models have extended their applications across diverse domains, including medicine, robotics, and content synthesis. These advanced GAN models have gained recognition for their excellent accuracy by scaling the model. However, existing accelerators face scalability challenges when dealing with large-scale GAN models. As the size of GAN models increases, the demand for computation and communication resources during inference continues to grow. To address this scalability issue, this article proposes Chiplet-GAN, a chiplet-based accelerator design for GAN inference. Chiplet-GAN enables scalability by adding more chiplets to the system, thereby supporting the scaling of computation capabilities. To handle the increasing communication demand as the system and model scale, a novel interconnection network with adaptive topology and passive/active network links is developed to provide adequate communication support for Chiplet-GAN. Coupled with workload partition and allocation algorithms, Chiplet-GAN reduces execution time and energy consumption for GAN inference workloads as both model and chiplet-system scales. Evaluation results using various GAN models show the effectiveness of Chiplet-GAN. On average, compared to GANAX, SpAtten, and Simba, the Chiplet-GAN reduces execution time and energy consumption by 34% and 21%, respectively. Furthermore, as the system scales for large-scale GAN model inference, Chiplet-GAN achieves reductions in execution time of up to 63% compared to the Simba, a chiplet-based accelerator. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email