Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PePNet: A Periodicity-Perceived Workload Prediction Network Supporting Rare Occurrence of Heavy Workload

Feiyi Chen, Zhen Qin, Hailiang Zhao, Shuiguang Deng

TL;DR

PePNet addresses cloud workload forecasting under heavy-tail bursts by jointly modeling short-term dynamics, long-term trends, and automatically detected periodic patterns. It introduces a Periodicity-Perceived Mechanism to detect and adaptively fuse periodic information and an Achilles' Heel Loss to focus training on under-fitting parts of the sequence, with theoretical guarantees on periodicity error bounds and an automatic hyperparameter method. The approach yields substantial improvements in both overall and heavy-load prediction accuracy across Alibaba2018, SMD, and Dinda datasets, while incurring only modest training/inference overhead. These contributions advance reliable SLA-aware workload provisioning in variable cloud environments by robustly capturing periodicity and mitigating data-imbalance effects.

Abstract

Cloud providers can greatly benefit from accurate workload prediction. However, the workload of cloud servers is highly variable, with occasional heavy workload bursts. This makes workload prediction challenging. There are mainly two categories of workload prediction methods: statistical methods and neural-network-based ones. The former ones rely on strong mathematical assumptions and have reported low accuracy when predicting highly variable workload. The latter ones offer higher overall accuracy, yet they are vulnerable to data imbalance between heavy workload and common one. This impairs the prediction accuracy of neural network-based models on heavy workload. Either the overall inaccuracy of statistic methods or the heavy-workload inaccuracy of neural-network-based models can cause service level agreement violations. Thus, we propose PePNet to improve overall especially heavy workload prediction accuracy. It has two distinctive characteristics: (i) A Periodicity-Perceived Mechanism to detect the existence of periodicity and the length of one period automatically, without any priori knowledge. Furthermore, it fuses periodic information adaptively, which is suitable for periodic, lax periodic and aperiodic time series. (ii) An Achilles' Heel Loss Function iteratively optimizing the most under-fitting part in predicting sequence for each step, which significantly improves the prediction accuracy of heavy load. Extensive experiments conducted on Alibaba2018, SMD dataset and Dinda's dataset demonstrate that PePNet improves MAPE for overall workload by 20.0% on average, compared with state-of-the-art methods. Especially, PePNet improves MAPE for heavy workload by 23.9% on average.

PePNet: A Periodicity-Perceived Workload Prediction Network Supporting Rare Occurrence of Heavy Workload

TL;DR

PePNet addresses cloud workload forecasting under heavy-tail bursts by jointly modeling short-term dynamics, long-term trends, and automatically detected periodic patterns. It introduces a Periodicity-Perceived Mechanism to detect and adaptively fuse periodic information and an Achilles' Heel Loss to focus training on under-fitting parts of the sequence, with theoretical guarantees on periodicity error bounds and an automatic hyperparameter method. The approach yields substantial improvements in both overall and heavy-load prediction accuracy across Alibaba2018, SMD, and Dinda datasets, while incurring only modest training/inference overhead. These contributions advance reliable SLA-aware workload provisioning in variable cloud environments by robustly capturing periodicity and mitigating data-imbalance effects.

Abstract

Cloud providers can greatly benefit from accurate workload prediction. However, the workload of cloud servers is highly variable, with occasional heavy workload bursts. This makes workload prediction challenging. There are mainly two categories of workload prediction methods: statistical methods and neural-network-based ones. The former ones rely on strong mathematical assumptions and have reported low accuracy when predicting highly variable workload. The latter ones offer higher overall accuracy, yet they are vulnerable to data imbalance between heavy workload and common one. This impairs the prediction accuracy of neural network-based models on heavy workload. Either the overall inaccuracy of statistic methods or the heavy-workload inaccuracy of neural-network-based models can cause service level agreement violations. Thus, we propose PePNet to improve overall especially heavy workload prediction accuracy. It has two distinctive characteristics: (i) A Periodicity-Perceived Mechanism to detect the existence of periodicity and the length of one period automatically, without any priori knowledge. Furthermore, it fuses periodic information adaptively, which is suitable for periodic, lax periodic and aperiodic time series. (ii) An Achilles' Heel Loss Function iteratively optimizing the most under-fitting part in predicting sequence for each step, which significantly improves the prediction accuracy of heavy load. Extensive experiments conducted on Alibaba2018, SMD dataset and Dinda's dataset demonstrate that PePNet improves MAPE for overall workload by 20.0% on average, compared with state-of-the-art methods. Especially, PePNet improves MAPE for heavy workload by 23.9% on average.
Paper Structure (17 sections, 19 equations, 4 figures, 7 tables, 1 algorithm)

This paper contains 17 sections, 19 equations, 4 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview of PePNet
  4. Periodicity-Perceived Mechanism
  5. Periodicity-Mining Module
  6. Periodicity-Fusing Module
  7. Achilles' Heel Loss Function
  8. Experiment
  9. Experiment Setup
  10. Prediction Accuracy
  11. Time Overhead
  12. Hyperparameter Sensitivity
  13. Ablation Experiment
  14. Related Work
  15. Workload and Time Series Prediction
  16. ...and 2 more sections

Figures (4)

  • Figure 1: (a)-(b) Overall workload and heavy workload prediction accuracy. (c)-(d) The fitting process of Achilles' Heel Loss Function. (c) In the first half of the process, the most under-fitting part is on the right. (d) After some training iterations, the prediction error on the right diminishes and the most under-fitting part shifts to the left. Then, Achilles' Heel Loss Function pays more attention to the left part in this phase.
  • Figure 2: The overview of PePNet.
  • Figure 3: Illustrations for data division and periodicity-perceived mechanism.
  • Figure 4: (a) The value of the y-axis is the original workload divided by the maximum workload of its machine. The wider the width of the violin in the figure is, the more workloads are distributed near this value. (b) The figure shows the average consuming time of one forward propagation and backpropagation for one sample. The labels on the histogram are 1000 times of the actual value. Besides, we use the shorthand of Informer, Autoformer and Reformer in the xticks. (c) The figure shows the overall and heavy-workload MAE of PePNet for different $\gamma$ on all the three datasets. In the legend, we use dataset name* to represent the heavy-workload MAE on the specific dataset, while using dataset name to represent the overall MAE on the dataset. (d)-(i) These figures show overall and heavy-workload MAE of PePNet on different datasets with different combination of the length of $X_{in}$ and $p$.