Hardware Resource Analysis in Distributed Training with Edge Devices
When training a deep learning model with distributed training, the hardware resource utilization of each device depends on the model structure and the number of devices used for training. Distributed training has recently been applied to edge computing. Since edge devices have hardware resource limi...
Main Authors: | , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2019-12-01
|
Series: | Electronics |
Subjects: | |
Online Access: | https://www.mdpi.com/2079-9292/9/1/28 |
_version_ | 1811187586048196608 |
---|---|
author | Sihyeong Park Jemin Lee Hyungshin Kim |
author_facet | Sihyeong Park Jemin Lee Hyungshin Kim |
author_sort | Sihyeong Park |
collection | DOAJ |
description | When training a deep learning model with distributed training, the hardware resource utilization of each device depends on the model structure and the number of devices used for training. Distributed training has recently been applied to edge computing. Since edge devices have hardware resource limitations such as memory, there is a need for training methods that use hardware resources efficiently. Previous research focused on reducing training time by optimizing the synchronization process between edge devices or by compressing the models. In this paper, we monitored hardware resource usage based on the number of layers and the batch size of the model during distributed training with edge devices. We analyzed memory usage and training time variability as the batch size and number of layers increased. Experimental results demonstrated that, the larger the batch size, the fewer synchronizations between devices, resulting in less accurate training. In the shallow model, training time increased as the number of devices used for training increased because the synchronization between devices took more time than the computation time of training. This paper finds that efficient use of hardware resources for distributed training requires selecting devices in the context of model complexity and that fewer layers and smaller batches are required for efficient hardware use. |
first_indexed | 2024-04-11T14:04:57Z |
format | Article |
id | doaj.art-8ca7b3ef761a48098ff02242778f47f9 |
institution | Directory Open Access Journal |
issn | 2079-9292 |
language | English |
last_indexed | 2024-04-11T14:04:57Z |
publishDate | 2019-12-01 |
publisher | MDPI AG |
record_format | Article |
series | Electronics |
spelling | doaj.art-8ca7b3ef761a48098ff02242778f47f92022-12-22T04:19:55ZengMDPI AGElectronics2079-92922019-12-01912810.3390/electronics9010028electronics9010028Hardware Resource Analysis in Distributed Training with Edge DevicesSihyeong Park0Jemin Lee1Hyungshin Kim2Department of Computer Science and Engineering, Chungnam National University, Daejeon 34134, KoreaFuture Computing Research Division, Artificial Intelligence Research Laboratory, Electronics and Telecommunications Research Institute (ETRI), Daejeon 34129, KoreaThe Division of Computer Convergence, Chungnam National University, Daejeon 34134, KoreaWhen training a deep learning model with distributed training, the hardware resource utilization of each device depends on the model structure and the number of devices used for training. Distributed training has recently been applied to edge computing. Since edge devices have hardware resource limitations such as memory, there is a need for training methods that use hardware resources efficiently. Previous research focused on reducing training time by optimizing the synchronization process between edge devices or by compressing the models. In this paper, we monitored hardware resource usage based on the number of layers and the batch size of the model during distributed training with edge devices. We analyzed memory usage and training time variability as the batch size and number of layers increased. Experimental results demonstrated that, the larger the batch size, the fewer synchronizations between devices, resulting in less accurate training. In the shallow model, training time increased as the number of devices used for training increased because the synchronization between devices took more time than the computation time of training. This paper finds that efficient use of hardware resources for distributed training requires selecting devices in the context of model complexity and that fewer layers and smaller batches are required for efficient hardware use.https://www.mdpi.com/2079-9292/9/1/28deep learningdistributed trainingedge computinginternet of thingsperformance monitoring |
spellingShingle | Sihyeong Park Jemin Lee Hyungshin Kim Hardware Resource Analysis in Distributed Training with Edge Devices Electronics deep learning distributed training edge computing internet of things performance monitoring |
title | Hardware Resource Analysis in Distributed Training with Edge Devices |
title_full | Hardware Resource Analysis in Distributed Training with Edge Devices |
title_fullStr | Hardware Resource Analysis in Distributed Training with Edge Devices |
title_full_unstemmed | Hardware Resource Analysis in Distributed Training with Edge Devices |
title_short | Hardware Resource Analysis in Distributed Training with Edge Devices |
title_sort | hardware resource analysis in distributed training with edge devices |
topic | deep learning distributed training edge computing internet of things performance monitoring |
url | https://www.mdpi.com/2079-9292/9/1/28 |
work_keys_str_mv | AT sihyeongpark hardwareresourceanalysisindistributedtrainingwithedgedevices AT jeminlee hardwareresourceanalysisindistributedtrainingwithedgedevices AT hyungshinkim hardwareresourceanalysisindistributedtrainingwithedgedevices |