Transformer-based latency prediction for stream processing task

Abstract Latency prediction for stream processing tasks (SPTs) is a critical issue for stream computing, parameter tuning, load optimization, task scheduling, etc. This study addresses the real-time, volatile, and high-volume nature of streaming workloads to improve latency prediction accuracy. A no...

Full description

Saved in:
Bibliographic Details
Main Authors: Zheng Chu, Baozhu Li, Changtian Ying
Format: Article
Language:English
Published: Springer 2025-07-01
Series:Journal of King Saud University: Computer and Information Sciences
Subjects:
Online Access:https://doi.org/10.1007/s44443-025-00089-0
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849331552344866816
author Zheng Chu
Baozhu Li
Changtian Ying
author_facet Zheng Chu
Baozhu Li
Changtian Ying
author_sort Zheng Chu
collection DOAJ
description Abstract Latency prediction for stream processing tasks (SPTs) is a critical issue for stream computing, parameter tuning, load optimization, task scheduling, etc. This study addresses the real-time, volatile, and high-volume nature of streaming workloads to improve latency prediction accuracy. A novel model based on Auto-encoders and Transformers was proposed to address the above challenges. The Auto-encoder is utilized to reduce the dimensionality of the extensive features and generate a compressed feature representation. Subsequently, the Transformer is employed to extract spatio-temporal dependencies and predict the latency of SPTs. Experimental results on six benchmarks (about 260,000 samples in total) demonstrate that compared to state-of-the-art models, the proposed model achieves higher accuracy and lower error rates, with average errors of 0.11% (MAE), 0.02% (RMSE), respectively, and average accuracy of 0.89. Moreover, a time window length of 60 or 80 allows the model to achieve the best average accuracy. Additionally, the ablation study verified that the Transformer significantly improved the prediction performance. These results confirm that the proposed model can accurately predict latency for SPTs, further laying the foundation for intelligent and proactive optimization of stream processing systems (SPSs).
format Article
id doaj-art-44371836e08043099a149eee59071e53
institution Kabale University
issn 1319-1578
2213-1248
language English
publishDate 2025-07-01
publisher Springer
record_format Article
series Journal of King Saud University: Computer and Information Sciences
spelling doaj-art-44371836e08043099a149eee59071e532025-08-20T03:46:29ZengSpringerJournal of King Saud University: Computer and Information Sciences1319-15782213-12482025-07-0137511810.1007/s44443-025-00089-0Transformer-based latency prediction for stream processing taskZheng Chu0Baozhu Li1Changtian Ying2School of Information and Electrical Engineering, Hangzhou City UniversityDonghai LaboratoryDepartment of Computers, Shaoxing UniversityAbstract Latency prediction for stream processing tasks (SPTs) is a critical issue for stream computing, parameter tuning, load optimization, task scheduling, etc. This study addresses the real-time, volatile, and high-volume nature of streaming workloads to improve latency prediction accuracy. A novel model based on Auto-encoders and Transformers was proposed to address the above challenges. The Auto-encoder is utilized to reduce the dimensionality of the extensive features and generate a compressed feature representation. Subsequently, the Transformer is employed to extract spatio-temporal dependencies and predict the latency of SPTs. Experimental results on six benchmarks (about 260,000 samples in total) demonstrate that compared to state-of-the-art models, the proposed model achieves higher accuracy and lower error rates, with average errors of 0.11% (MAE), 0.02% (RMSE), respectively, and average accuracy of 0.89. Moreover, a time window length of 60 or 80 allows the model to achieve the best average accuracy. Additionally, the ablation study verified that the Transformer significantly improved the prediction performance. These results confirm that the proposed model can accurately predict latency for SPTs, further laying the foundation for intelligent and proactive optimization of stream processing systems (SPSs).https://doi.org/10.1007/s44443-025-00089-0Streaming dataStream processing taskLatency predictionTransformer
spellingShingle Zheng Chu
Baozhu Li
Changtian Ying
Transformer-based latency prediction for stream processing task
Journal of King Saud University: Computer and Information Sciences
Streaming data
Stream processing task
Latency prediction
Transformer
title Transformer-based latency prediction for stream processing task
title_full Transformer-based latency prediction for stream processing task
title_fullStr Transformer-based latency prediction for stream processing task
title_full_unstemmed Transformer-based latency prediction for stream processing task
title_short Transformer-based latency prediction for stream processing task
title_sort transformer based latency prediction for stream processing task
topic Streaming data
Stream processing task
Latency prediction
Transformer
url https://doi.org/10.1007/s44443-025-00089-0
work_keys_str_mv AT zhengchu transformerbasedlatencypredictionforstreamprocessingtask
AT baozhuli transformerbasedlatencypredictionforstreamprocessingtask
AT changtianying transformerbasedlatencypredictionforstreamprocessingtask