pith. sign in

arxiv: 2606.09787 · v1 · pith:3TYEIUTG · submitted 2026-06-08 · cs.LG · cs.NI

Zero Touch Predictive Orchestration: Automating Time-Series Models for the Cloud-Edge Continuum

Abd Elghani Meliani , Arora Sagar , Adlen Ksentini , Raymond Knopp This is my paper

pith:3TYEIUTGreviewed 2026-06-27 17:19 UTCmodel grok-4.3open to challenge →

classification cs.LG cs.NI
keywords cold starttime-series forecastingcloud-edge continuumzero touch managementdata mixingneural architecture searchtelemetryorchestration
0
0 comments X

The pith

Merging sparse local samples with the TimeTrack dataset overcomes the cold-start problem for time-series forecasting on new cloud-edge nodes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper targets the cold-start issue in zero-touch orchestration for the Cloud-Edge Continuum, where newly discovered nodes have too few telemetry samples to train reliable predictive models. It introduces an automated pipeline that collects local data via a Resource Exposer and mixes it with TimeTrack, a high-resolution public dataset, before feeding the combination into a Neural Architecture Search engine. The resulting models are claimed to deliver lower error rates and faster training than local-only data, generic datasets, or other mixtures. A sympathetic reader would care because successful data synergy would let orchestrators deploy accurate forecasts immediately on volatile edge hardware without waiting for long local histories.

Core claim

The paper claims that automatically merging target node data with TimeTrack mitigates the cold start challenge. This integration significantly improves forecasting accuracy measured in Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) and accelerates convergence compared to training on the sparse local samples alone, training solely on generic datasets, or mixing the target data with standard alternative datasets.

What carries the argument

The data-mixing methodology that combines sparse local telemetry with TimeTrack to transfer high-frequency temporal patterns while calibrating to node-specific behaviors, followed by automated Neural Architecture Search for model generation.

If this is right

  • Forecasting accuracy improves on the three standard error metrics when the mixed dataset is used.
  • Model training reaches target performance in fewer epochs than the compared baselines.
  • The resulting models supply a stable starting point for ongoing MLOps retraining loops.
  • Proactive zero-touch management becomes feasible even for nodes that have just been discovered.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same mixing step could shorten the data-gathering window required before an edge node can run its own predictive controller.
  • If the synergy holds across hardware types, the method might reduce reliance on long-running local monitors in other distributed systems that face cold-start forecasting.
  • One could test whether the accuracy gain persists when the local samples come from different telemetry granularities or from entirely different application domains.

Load-bearing premise

The high-frequency patterns recorded in TimeTrack remain useful and non-conflicting when added to the unique hardware and microservice traces of any newly discovered node.

What would settle it

A side-by-side experiment in which models trained on the TimeTrack-plus-local mixture show equal or higher MSE, MAE, or MAPE than models trained on local samples alone or on local samples mixed with other public datasets would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.09787 by Abd Elghani Meliani, Adlen Ksentini, Arora Sagar, Raymond Knopp.

Figure 1
Figure 1. Figure 1: High-level architecture of the proactive resource forecasting system presented in [3]. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example of critical short-term patterns that are smoothed over and lost due to coarse 5-minute data collection [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed solution architecture for CEC environments. The first two layers (Data Acquisition [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: End-to-end architecture of the automated forecasting pipeline, illustrating the flow from System Discovery [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The lower-level architecture of the Resource Exposer, illustrating the plugin-based data collection and message [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Interaction workflow between the different framework components, highlighting registration, continuous [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Correlation matrix demonstrating relationships between CPU usage (CU), memory usage (UM), disk [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Memory utilization across the cluster over a week, showing distinct usage spikes during weekdays versus low, [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: CPU consumption (left) and memory usage (right) of the Resource Exposer and the internal broker across [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of API response times under concurrent request loads, demonstrating the efficiency of the RE’s [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Performance heatmap of NAS CPU models comparing error metrics (MAPE, MAE, MSE) across varying [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Impact of target data size (10 to 500 samples) on the minimum achieved MAPE across isolated and mixed [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Minimum prediction error (MAPE) across different forecasting horizons (from 1 to 5 steps ahead) for all [PITH_FULL_IMAGE:figures/full_fig_p015_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Training time, memory, and CPU usage for deployed models. The graphs scale proportionally to represent [PITH_FULL_IMAGE:figures/full_fig_p017_14.png] view at source ↗
read the original abstract

The Cloud-Edge Continuum (CEC) enables latency-critical applications by distributing resources to the far edge, but its extreme volatility makes proactive Zero Touch Management via time-series forecasting essential. However, orchestrators face a severe "cold start" problem: newly discovered nodes lack the historical data required to train localized predictive models, while generalized models fail to capture unique hardware and microservice behaviors. To solve this, we propose a fully automated time-series prediction architecture driven by a novel data-mixing methodology. At the infrastructure level, we introduce a lightweight, technology-agnostic Resource Exposer (RE) that dynamically discovers nodes and continuously collects customizable telemetry (e.g., compute, network, energy). To overcome the sparsity of these initial local samples, our framework automatically merges them with TimeTrack, our publicly available, high-resolution dataset collected at 45-second intervals. This synergizes TimeTrack's foundational, high-frequency temporal patterns with the precise calibration of the local node data. Processed through a Neural Architecture Search (NAS) engine, the system automatically generates highly accurate baseline models. Experimental results demonstrate that merging the target data with TimeTrack effectively mitigates the cold start challenge. This integration significantly improves forecasting accuracy measured in Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) and accelerates convergence compared to training on the sparse local samples alone, training solely on generic datasets, or mixing the target data with standard alternative datasets, establishing a robust foundation for continuous MLOps deployment.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper claims to address the cold-start problem in time-series forecasting for Cloud-Edge Continuum orchestration by introducing a lightweight Resource Exposer (RE) for dynamic telemetry collection and a data-mixing strategy that combines sparse local node samples with the high-resolution TimeTrack dataset. This mixed data is fed to a Neural Architecture Search (NAS) engine to automatically produce baseline models, with the abstract asserting that the approach yields measurable gains in MSE, MAE, and MAPE plus faster convergence relative to local-only training, generic datasets, or mixing with other standard datasets.

Significance. If the claimed empirical gains are substantiated with full experimental details, the work could offer a practical, automated pathway for zero-touch predictive management in volatile edge settings, reducing reliance on manual model tuning and supporting continuous MLOps. The public availability of TimeTrack is a constructive element that could aid reproducibility if the mixing protocol is fully specified.

major comments (2)
  1. [Abstract] Abstract: The central empirical claim—that mixing with TimeTrack produces significant improvements in MSE/MAE/MAPE and convergence—is asserted without any numerical results, error bars, dataset statistics, baseline values, or statistical tests, rendering the outcome unverifiable and load-bearing for the paper's contribution.
  2. [Abstract] Abstract: No description is given of the data-mixing procedure (e.g., temporal alignment, mixing ratios, preprocessing), the NAS search space or objective, or the underlying time-series model family, all of which are required to assess whether the reported synergy is reproducible or an artifact of unspecified implementation choices.
minor comments (1)
  1. [Abstract] The acronym CEC is introduced without an explicit expansion on first use.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We agree that it should be more self-contained with quantitative evidence and methodological outlines to support verifiability, and we will revise accordingly while preserving the paper's core claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central empirical claim—that mixing with TimeTrack produces significant improvements in MSE/MAE/MAPE and convergence—is asserted without any numerical results, error bars, dataset statistics, baseline values, or statistical tests, rendering the outcome unverifiable and load-bearing for the paper's contribution.

    Authors: We agree the abstract should include concrete numerical support for the claimed gains. The full manuscript reports these results (with error bars and comparisons) in the experimental evaluation, but the abstract will be updated to state specific improvements (e.g., relative reductions in MSE/MAE/MAPE and faster convergence epochs) and reference the statistical tests performed. This addresses verifiability without changing the underlying experiments. revision: yes

  2. Referee: [Abstract] Abstract: No description is given of the data-mixing procedure (e.g., temporal alignment, mixing ratios, preprocessing), the NAS search space or objective, or the underlying time-series model family, all of which are required to assess whether the reported synergy is reproducible or an artifact of unspecified implementation choices.

    Authors: We concur that a brief methodological sketch belongs in the abstract. The revised abstract will concisely describe the mixing approach (including alignment and ratios), the NAS search space/objective, and the base model family, while directing readers to the detailed methodology sections for full reproducibility. These elements are already specified in the body of the paper. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper presents an empirical claim that mixing sparse local telemetry with the authors' TimeTrack dataset improves MSE/MAE/MAPE and convergence speed versus local-only, generic-only, or other-mix baselines. No equations, derivations, fitted parameters renamed as predictions, or self-referential definitions appear in the provided text. The central result is framed as an experimental outcome of data mixing rather than a mathematical reduction. No load-bearing self-citation chains, uniqueness theorems, or ansatzes are invoked to force the result. The derivation chain is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim depends on the transferability of patterns from TimeTrack to new nodes and on the effectiveness of the mixing procedure; these are domain assumptions not supported by independent evidence in the abstract. No free parameters or invented entities with external falsifiability are detailed.

axioms (1)
  • domain assumption TimeTrack dataset contains foundational high-frequency temporal patterns that synergize with local node data to improve forecasts.
    Directly invoked to justify the data-mixing methodology as overcoming sparsity.
invented entities (1)
  • Resource Exposer (RE) no independent evidence
    purpose: Lightweight component to dynamically discover nodes and collect customizable telemetry.
    Introduced as the infrastructure-level data collection mechanism.

pith-pipeline@v0.9.1-grok · 5818 in / 1274 out tokens · 20910 ms · 2026-06-27T17:19:39.290985+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

30 extracted references

  1. [1]

    Timetrack: A dataset for exploring temporal patterns and predictive insights into openairinterface (oai) ci/cd cluster

    Meliani Abd Elghani, Sagar Arora, Adlen Ksentini, and Raymond Knopp. Timetrack: A dataset for exploring temporal patterns and predictive insights into openairinterface (oai) ci/cd cluster. In IEEE, editor,ICC 2025, IEEE International Conference on Communications 2025, 8-12 June 2025, Montreal, Canada, Montreal, 2025

    2025

  2. [2]

    Ai-native cecc management architecture: Enabling scalable and intelligent cloud-edge computing

    Abd Elghani Meliani, Adlen Ksentini, Mohamed Mekki, Abdelhak Kadouma, Dimitrios Amaxilatis, Amadou Ba, Eduardo Ojeda Coronado, John Beredimas, Vrettos Mouloss, Souvik Sengupta, Dimitris Klonidis, and Christos Verikoukis. Ai-native cecc management architecture: Enabling scalable and intelligent cloud-edge computing. Athens, 2025

    2025

  3. [3]

    Resiliency focused proactive lifecycle management for stateful microservices in multi-cluster containerized environments.Computer Communications, 236:108111, 2025

    Abd Elghani Meliani, Mohamed Mekki, and Adlen Ksentini. Resiliency focused proactive lifecycle management for stateful microservices in multi-cluster containerized environments.Computer Communications, 236:108111, 2025

    2025

  4. [4]

    Layer-reuse aware optimization for efficient microservice migration in uav edge systems

    Miloud Bagaa, Adlen Ksentini, et al. Layer-reuse aware optimization for efficient microservice migration in uav edge systems. InICC 2026, IEEE International Conference on Communications, 2026

    2026

  5. [5]

    Lightweight resource exposure framework for efficient service and resource orchestration in the cloud-edge continuum

    Abd Elghani Meliani and Adlen Ksentini. Lightweight resource exposure framework for efficient service and resource orchestration in the cloud-edge continuum. In IEEE, editor,ICC 2025, IEEE International Conference on Communications, 2nd Workshop on the Path Towards 6G: Standardization and Research Vision, 8-12 June 2025, Montreal, Canada, Montreal, 2025

    2025

  6. [6]

    OpenAirInterface: 5G Software Alliance for Democratising Wireless Innovation, 2023

    OpenAirInterface. OpenAirInterface: 5G Software Alliance for Democratising Wireless Innovation, 2023. Accessed: 2024-10-30

    2023

  7. [7]

    OpenAirInterface Ci/Cd webpage, 2023

    OpenAirInterface. OpenAirInterface Ci/Cd webpage, 2023. Accessed: 2024-10-30

    2023

  8. [8]

    A.J. Jerri. The shannon sampling theorem—its various extensions and applications: A tutorial review.Proceedings of the IEEE, 65(11):1565–1596, 1977. 18 APREPRINT- JUNE9, 2026

    1977

  9. [9]

    A review of neural architecture search.Neurocomput- ing, 474:82–93, 2022

    Dilyara Baymurzina, Eugene Golikov, and Mikhail Burtsev. A review of neural architecture search.Neurocomput- ing, 474:82–93, 2022

    2022

  10. [10]

    Timetrack: Openairinterface (oai) ci/cd time series dataset, 2025

    Abdelghani Meliani. Timetrack: Openairinterface (oai) ci/cd time series dataset, 2025. Accessed: Feb 13, 2025

    2025

  11. [11]

    Timetrack: Openairinterface (oai) ci/cd time series dataset, 2025

    Abdelghani Meliani. Timetrack: Openairinterface (oai) ci/cd time series dataset, 2025. Accessed: Aug 26, 2025

    2025

  12. [12]

    Grid Workloads Archive GW A-T-13-Materna

    Delft University of Technology. Grid Workloads Archive GW A-T-13-Materna. Accessed: 2024-08-27

    2024

  13. [13]

    Google Cluster Data 2019, 2019

    Google Inc. Google Cluster Data 2019, 2019. Accessed: 2024-08-27

    2019

  14. [14]

    Microsoft Azure Public Dataset V1, 2019

    Microsoft Corporation. Microsoft Azure Public Dataset V1, 2019. Accessed: 2024-10-30

    2019

  15. [15]

    Alibaba Cluster Trace v2018, 2018

    Alibaba Inc. Alibaba Cluster Trace v2018, 2018. Accessed: 2024-08-27

    2018

  16. [16]

    Microsoft Azure Public Dataset V2, 2020

    Microsoft Corporation. Microsoft Azure Public Dataset V2, 2020. Accessed: 2024-10-30

    2020

  17. [17]

    Mai et al

    J. Mai et al. Is sampled data sufficient for anomaly detection? InProceedings of IMC, 2006

    2006

  18. [18]

    PhD thesis, ETH Zurich, 2019

    Antonio Libri.Towards High-Resolution Monitoring for HPC and Data Center Analytics, Automation and Control. PhD thesis, ETH Zurich, 2019

    2019

  19. [19]

    Effects of resampled data on time series forecasting accuracy

    Jennifer Garland. Effects of resampled data on time series forecasting accuracy. 2019

    2019

  20. [20]

    Deep learning.Nature, 521(7553):436–444, May 2015

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning.Nature, 521(7553):436–444, May 2015

    2015

  21. [21]

    Long short-term memory.Neural Computation, 9(8):1735–1780, 11 1997

    Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory.Neural Computation, 9(8):1735–1780, 11 1997

    1997

  22. [22]

    What is mlops? InBeginning MLOps with MLFlow: Deploy Models in A WS SageMaker , Google Cloud, and Microsoft Azure, pages 79–124

    Sridhar Alla and Suman Kalyan Adari. What is mlops? InBeginning MLOps with MLFlow: Deploy Models in A WS SageMaker , Google Cloud, and Microsoft Azure, pages 79–124. Springer, 2020

    2020

  23. [23]

    Automl to date and beyond: Challenges and opportunities.Acm computing surveys (csur), 54(8):1–36, 2021

    Shubhra Kanti Karmaker, Md Mahadi Hassan, Micah J Smith, Lei Xu, Chengxiang Zhai, and Kalyan Veeramacha- neni. Automl to date and beyond: Challenges and opportunities.Acm computing surveys (csur), 54(8):1–36, 2021

    2021

  24. [24]

    Decor: A distributed coordinated resource monitoring system

    Shan-Hsiang Shen and Aditya Akella. Decor: A distributed coordinated resource monitoring system. In2012 IEEE 20th International Workshop on Quality of Service, pages 1–9, 2012

    2012

  25. [25]

    Dprof-distributed system profiling and tracing

    Tuan Nguyen and Manish Pandey. Dprof-distributed system profiling and tracing

  26. [26]

    A scalable monitoring framework for network slicing in 5g and beyond mobile networks.IEEE Transactions on Network and Service Management, 19(1):413–423, 2022

    Mohamed Mekki, Sagar Arora, and Adlen Ksentini. A scalable monitoring framework for network slicing in 5g and beyond mobile networks.IEEE Transactions on Network and Service Management, 19(1):413–423, 2022

    2022

  27. [27]

    Contreras, Jordi Ros-Giralt, and Roland Schott

    Sabine Randriamasy, Luis M. Contreras, Jordi Ros-Giralt, and Roland Schott. Operational compute metrics, 2024-07-07

    2024

  28. [28]

    Prometheus: Monitoring system and time series database, n.d

    Prometheus. Prometheus: Monitoring system and time series database, n.d

  29. [29]

    Nagios - the industry standard in it infrastructure monitoring, 2024

    Nagios. Nagios - the industry standard in it infrastructure monitoring, 2024. Accessed: 2024-09-12

    2024

  30. [30]

    Neural Network Intelligence, 1 2021

    Microsoft. Neural Network Intelligence, 1 2021. 19

    2021