Pith. sign in

REVIEW 2 major objections 54 references

A linear graph transformer on span graphs predicts microservice tail latencies more accurately and with faster inference than prior graph neural networks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-05-07 13:29 UTC

load-bearing objection STLGT applies a linear graph transformer to span graphs from microservice traces for tail latency prediction and reports solid gains over PERT-GNN, but the N=32 cap on Alibaba traces undercuts the long-range claim. the 2 major comments →

arxiv 2604.26422 v1 submitted 2026-04-29 cs.LG cs.AI

STLGT: A Scalable Trace-Based Linear Graph Transformer for Tail Latency Prediction in Microservices

Yongliang Ding , Qigong Bi , Peng Pu This is my paper
classification cs.LG cs.AI
keywords tail latency predictionmicroservicesgraph transformerspan graphsSLO managementtrace analysisworkload forecastingperformance prediction
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces STLGT to forecast end-to-end tail latency in microservice systems by encoding traces as span graphs. It applies a structure-aware linear graph transformer to propagate cross-service dependencies with inference time that scales linearly with graph size, plus a separate temporal module to handle workload changes. Tests across a personalized education application, DeathStarBench, and Alibaba traces show an average 8.5 percent MAPE improvement over PERT-GNN along with up to 12 times faster CPU inference at the largest tested graph size. The work targets proactive service level objective management by addressing long-range dependencies and bursty traffic patterns. If the gains hold, operators could rely on these forecasts to adjust resources before latency violations occur.

Core claim

STLGT is a per-API predictor that represents traces as span graphs, uses a structure-aware linear graph Transformer to propagate cross-service dependencies with linear inference time, and adds a decoupled temporal module to capture non-stationary workload dynamics for accurate multi-step p95 tail-latency forecasting.

What carries the argument

Structure-aware linear graph Transformer for propagating dependencies across span graphs, combined with a decoupled temporal module that isolates workload dynamics.

Load-bearing premise

Representing traces as span graphs and routing information through the linear graph transformer plus temporal module will capture essential long-range dependencies and workload effects without substantial information loss or overfitting to the evaluated traces.

What would settle it

A new collection of microservice traces containing longer dependency chains or burst patterns absent from the training set where the MAPE gain over PERT-GNN vanishes or inference time ceases to scale linearly with graph size.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • More accurate tail-latency forecasts allow earlier and more precise adjustments to meet service level objectives in production microservices.
  • Linear inference time supports processing of larger trace graphs without proportional slowdowns after preprocessing.
  • The decoupled temporal module improves handling of non-stationary and bursty traffic compared to models that entangle structure and time.
  • Consistent gains across synthetic benchmarks and real Alibaba traces suggest the method generalizes beyond the specific test applications.

Where Pith is reading between the lines

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

  • If span graphs preserve the key causal structure, the same linear transformer could be applied to latency prediction in other distributed systems such as cloud workflows or serverless functions.
  • Faster CPU inference opens the possibility of embedding the model in online monitoring pipelines that update predictions every few seconds.
  • The separation of structural and temporal modules might simplify extensions to multi-objective optimization that jointly predict latency and resource usage.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 0 minor

Summary. The paper introduces STLGT, a per-API tail-latency predictor that encodes microservice traces as span graphs, applies a structure-aware linear graph transformer to propagate cross-service dependencies in linear time, and uses a decoupled temporal module to handle workload dynamics. It claims an average 8.5% MAPE improvement over PERT-GNN across a personalized education microservice, DeathStarBench, and Alibaba traces, plus up to 12x faster CPU inference at N=32 (the maximum span-graph size after Alibaba preprocessing), with ablations confirming component value under bursty traffic.

Significance. If the accuracy and scalability results hold under rigorous validation, the work offers a practical advance for proactive SLO management in microservices by combining graph-structured dependency modeling with efficient linear attention. The explicit linear complexity in span-graph size and focus on non-stationary workloads address real deployment constraints; reproducible code or detailed protocols would further strengthen its contribution.

major comments (2)
  1. [Abstract] Abstract: the claim that STLGT reliably captures long-range dependency propagation and non-stationary bursty effects rests on the span-graph representation plus structure-aware linear transformer. However, the abstract states that N=32 is the maximum span-graph size 'after preprocessing the Alibaba traces.' If this preprocessing truncates or prunes longer traces (common in bursty workloads), the input already discards the very long-range effects the decoupled temporal module and linear attention are designed to model. This makes the reported 8.5% MAPE gain and 'scalable' positioning potentially specific to the capped subset rather than a general demonstration of long-range modeling.
  2. [Experimental Evaluation] Experimental section (implied by ablation and comparison results): the concrete MAPE and speedup numbers are reported without error bars, exact train/test splits, hyperparameter settings, or statistical significance tests. This absence undermines confidence in the 8.5% average improvement and 12x speedup claims, as it prevents assessment of variability across runs or traces.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major point below, providing clarifications and indicating the revisions we will incorporate to strengthen the paper.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that STLGT reliably captures long-range dependency propagation and non-stationary bursty effects rests on the span-graph representation plus structure-aware linear transformer. However, the abstract states that N=32 is the maximum span-graph size 'after preprocessing the Alibaba traces.' If this preprocessing truncates or prunes longer traces (common in bursty workloads), the input already discards the very long-range effects the decoupled temporal module and linear attention are designed to model. This makes the reported 8.5% MAPE gain and 'scalable' positioning potentially specific to the capped subset rather than a general demonstration of long-range modeling.

    Authors: We appreciate this observation on the preprocessing step. The Alibaba trace preprocessing consists of standard filtering to remove invalid spans (e.g., those with missing timestamps or protocol errors) while preserving all dependency chains present in the original traces; the resulting maximum span-graph size is 32. Because our structure-aware linear attention has O(N) complexity, the model can directly handle larger graphs when they appear in other workloads. To eliminate any ambiguity, we will revise the abstract and add an explicit paragraph in Section 4.1 describing the preprocessing pipeline and confirming that long-range dependencies within each trace are retained. revision: partial

  2. Referee: [Experimental Evaluation] Experimental section (implied by ablation and comparison results): the concrete MAPE and speedup numbers are reported without error bars, exact train/test splits, hyperparameter settings, or statistical significance tests. This absence undermines confidence in the 8.5% average improvement and 12x speedup claims, as it prevents assessment of variability across runs or traces.

    Authors: We agree that these details are necessary for rigorous evaluation. In the revised manuscript we will: (i) report mean and standard deviation of MAPE and inference time over five independent runs with different random seeds, (ii) specify the exact temporal train/validation/test splits (70/15/15) used to prevent leakage, (iii) list all hyperparameter values and the grid-search procedure, and (iv) include paired t-test p-values comparing STLGT against each baseline. These additions will allow readers to assess both variability and statistical significance of the reported improvements. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation or performance claims

full rationale

The paper presents an empirical architecture (structure-aware linear graph transformer plus decoupled temporal module) evaluated on preprocessed traces from multiple benchmarks. No equations, derivations, or self-citations are shown that reduce the reported MAPE gains or inference speedups to fitted parameters or inputs by construction. The N=32 span-graph cap is a stated preprocessing choice for the Alibaba traces, not a self-referential step that forces the accuracy results. Ablation studies and comparisons to PERT-GNN remain independent of any internal loop.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on standard assumptions that graph neural networks can propagate service dependencies and that linear attention approximations preserve accuracy for this domain; no explicit free parameters or invented physical entities are stated in the abstract.

pith-pipeline@v0.9.0 · 5467 in / 1178 out tokens · 65830 ms · 2026-05-07T13:29:21.997787+00:00 · methodology

0 comments
read the original abstract

Accurate end-to-end tail-latency forecasting is critical for proactive SLO management in microservice systems. However, modeling long-range dependency propagation and non-stationary, bursty workloads while maintaining inference efficiency at scale remains challenging. We present STLGT (Scalable Trace-based Linear Graph Transformer), a per-API predictor that encodes traces as span graphs for multi-step p95 tail-latency forecasting. STLGT uses a structure-aware linear graph Transformer to propagate cross-service dependencies with inference time linear in span graph size, and a decoupled temporal module to capture workload dynamics. Across a personalized education microservice application, DeathStarBench, and Alibaba traces, STLGT improves forecasting accuracy over PERT-GNN by 8.5% MAPE on average and achieves up to 12x faster CPU inference at N=32, matching the maximum span graph size after preprocessing the Alibaba traces. Ablation studies further demonstrate the effectiveness of each component, especially under bursty traffic.

Figures

Figures reproduced from arXiv: 2604.26422 by Peng Pu, Qigong Bi, Yongliang Ding.

Figure 1
Figure 1. Figure 1: Challenges in Modeling Long-Chain Dependencies view at source ↗
Figure 2
Figure 2. Figure 2: The overall architecture of STLGT nodes and E A encodes their invocation dependencies. Each node 𝑣 A 𝑖 ∈ VA corresponds to a span graph representation of some microservice 𝑠 ∈ V. We define a surjective mapping 𝜋 : VA → V that maps each stage node to its corresponding microservice. Since a microservice may be invoked multiple times within a trace, multiple nodes in VA may map to the same microservice, and t… view at source ↗
Figure 3
Figure 3. Figure 3: Spatial encoder architecture of STLGT with the Structure-aware linear attention module. view at source ↗
Figure 4
Figure 4. Figure 4: Workload and production-trace characteristics. (a) view at source ↗
Figure 5
Figure 5. Figure 5: Effect of hidden width 𝑑 on mean MAPE: (a) DSB microservice benchmarks, including Social Network and Ho￾tel Reservation; (b) Edu Platform. Lower is better. Across both benchmark and education settings, 𝑑 = 16 is too narrow and leads to noticeably higher error, indicating that the model needs sufficient capacity to encode dependency propagation and temporal dynamics. Moving to a moderate width (𝑑 = 32 or 64… view at source ↗

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

54 extracted references · 54 canonical work pages · 1 internal anchor

  1. [1]

    Ataollah Fatahi Baarzi and George Kesidis. 2021. Showar: Right-sizing and efficient scheduling of microservices. InProceedings of the ACM Symposium on Cloud Computing. Association for Computing Machinery, New York, NY, USA, 427–441

    work page 2021

  2. [2]

    Iz Beltagy, Matthew E Peters, and Arman Cohan. 2020. Longformer: The long- document transformer. arXiv preprint arXiv:2004.05150

    work page internal anchor Pith review arXiv 2020

  3. [3]

    Jiabin Chen, Fei Xu, Yikun Gu, Li Chen, Fangming Liu, and Zhi Zhou. 2024. Har- monyBatch: Batching multi-SLO DNN inference with heterogeneous serverless functions. In2024 IEEE/ACM 32nd International Symposium on Quality of Service (IWQoS). IEEE, IEEE, Piscataway, NJ, USA, 1–10

    work page 2024

  4. [4]

    Tao Chen, Rami Bahsoon, and Xin Yao. 2018. A Survey and Taxonomy of Self- Aware and Self-Adaptive Cloud Autoscaling Systems.Comput. Surveys51, 3 (June 2018), 1–40. doi:10.1145/3190507

    work page doi:10.1145/3190507 2018

  5. [5]

    Thomas Davidson, Emily Wall, and Jonathan Mace. 2023. A qualitative interview study of distributed tracing visualisation: A characterisation of challenges and opportunities.IEEE Transactions on Visualization and Computer Graphics30, 7 (2023), 3828–3840

    work page 2023

  6. [6]

    Shuiguang Deng, Hailiang Zhao, Binbin Huang, Cheng Zhang, Feiyi Chen, Yinuo Deng, Jianwei Yin, Schahram Dustdar, and Albert Y Zomaya. 2024. Cloud-native computing: A survey from the perspective of services.Proc. IEEE112, 1 (2024), 12–46

    work page 2024

  7. [7]

    Nicola Dragoni, Saverio Giallorenzo, Alberto Lluch Lafuente, Manuel Mazzara, Fabrizio Montesi, Ruslan Mustafin, and Larisa Safina. 2017. Microservices: yester- day, today, and tomorrow. InPresent and Ulterior Software Engineering. Springer International Publishing, Cham, Switzerland, 195–216

    work page 2017

  8. [8]

    Fanrong Du, Jiuchen Shi, Quan Chen, Pu Pang, Li Li, and Minyi Guo. 2025. Generating Microservice Graphs with Production Characteristics for Efficient Resource Scaling. InProceedings of the 39th ACM International Conference on Supercomputing. Association for Computing Machinery, New York, NY, USA, 895–910

    work page 2025

  9. [9]

    Jerome H Friedman. 2001. Greedy function approximation: a gradient boosting machine.The Annals of Statistics29, 5 (2001), 1189–1232

    work page 2001

  10. [10]

    Yu Gan et al. 2019. An open-source benchmark suite for microservices and their hardware-software implications for cloud & edge systems. InProceedings of the 24th International Conference on Architectural Support for Programming Languages and Operating Systems. Association for Computing Machinery, New York, NY, USA, 3–18

    work page 2019

  11. [11]

    Yu Gan, Yanqi Zhang, Kelvin Hu, Dailun Cheng, Yuan He, Meghna Pancholi, and Christina Delimitrou. 2019. Seer: Leveraging big data to navigate the complexity of performance debugging in cloud microservices. InProceedings of the twenty- fourth international conference on architectural support for programming languages and operating systems. Association for ...

    work page 2019

  12. [12]

    Alim Ul Gias, Giuliano Casale, and Murray Woodside. 2019. ATOM: Model- driven autoscaling for microservices. In2019 IEEE 39th International Conference on Distributed Computing Systems (ICDCS). IEEE, IEEE, Piscataway, NJ, USA, 1994–2004

    work page 2019

  13. [13]

    Weizhe Hua, Zihang Dai, Hanxiao Liu, and Quoc Le. 2022. Transformer quality in linear time. InInternational conference on machine learning. PMLR, PMLR, Baltimore, MD, USA, 9099–9117

    work page 2022

  14. [14]

    Gabriela Kadlecová, Jovita Lukasik, Martin Pilát, Petra Vidnerová, Mahmoud Safari, Roman Neruda, and Frank Hutter. 2024. Surprisingly strong performance prediction with neural graph features. arXiv preprint arXiv:2404.16551

    work page arXiv 2024

  15. [15]

    Angelos Katharopoulos, Apoorv Vyas, Nikolaos Pappas, and François Fleuret

    work page

  16. [16]

    InInternational conference on machine learning

    Transformers are rnns: Fast autoregressive transformers with linear at- tention. InInternational conference on machine learning. PMLR, PMLR, Virtual Event, 5156–5165

    work page

  17. [17]

    Leonardo Leite, Carla Rocha, Fabio Kon, Dejan Milojicic, and Paulo Meirelles

    work page

  18. [18]

    Surveys52, 6 (Nov

    A Survey of DevOps Concepts and Challenges.Comput. Surveys52, 6 (Nov. 2019), 1–35. doi:10.1145/3359981

    work page doi:10.1145/3359981 2019

  19. [19]

    Shutian Luo, Huanle Xu, Chengzhi Lu, Kejiang Ye, Guoyao Xu, Liping Zhang, Yu Ding, Jian He, and Chengzhong Xu. 2021. Characterizing microservice de- pendency and performance: Alibaba trace analysis. InProceedings of the ACM Symposium on Cloud Computing. Association for Computing Machinery, New York, NY, USA, 412–426

    work page 2021

  20. [20]

    Yang Luo, Mohan Gao, Zhemeng Yu, Haoyuan Ge, Xiaofeng Gao, Tengwei Cai, and Guihai Chen. 2024. Integrating system state into spatio temporal graph neural network for microservice workload prediction. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, New York, NY, USA, 5521–5531. EA...

    work page 2024

  21. [21]

    Ashraf Mahgoub, Edgardo Barsallo Yi, Karthick Shankar, Sameh Elnikety, Somali Chaterji, and Saurabh Bagchi. 2022. {ORION} and the three rights: Sizing, bundling, and prewarming for serverless {DAGs}. In16th USENIX Symposium on Operating Systems Design and Implementation (OSDI 22). USENIX Association, Berkeley, CA, USA, 303–320

    work page 2022

  22. [22]

    Chunyang Meng, Shijie Song, Haogang Tong, Maolin Pan, and Yang Yu. 2023. Deepscaler: Holistic autoscaling for microservices based on spatiotemporal gnn with adaptive graph learning. In2023 38th IEEE/ACM International Conference on Automated Software Engineering (ASE). IEEE, IEEE, Piscataway, NJ, USA, 53–65

    work page 2023

  23. [23]

    Amirhossein Mirhosseini, Sameh Elnikety, and Thomas F Wenisch. 2021. Parslo: A gradient descent-based approach for near-optimal partial slo allotment in microservices. InProceedings of the ACM Symposium on Cloud Computing. Asso- ciation for Computing Machinery, New York, NY, USA, 442–457

    work page 2021

  24. [24]

    Hoa X Nguyen, Shaoshu Zhu, and Mingming Liu. 2022. Graph-PHPA: graph- based proactive horizontal pod autoscaling for microservices using LSTM-GNN. In2022 IEEE 11th International Conference on Cloud Networking (CloudNet). IEEE, IEEE, Piscataway, NJ, USA, 237–241

    work page 2022

  25. [25]

    Jinwoo Park, Byungkwon Choi, Chunghan Lee, and Dongsu Han. 2021. GRAF: A graph neural network based proactive resource allocation framework for SLO- oriented microservices. InProceedings of the 17th International Conference on emerging Networking EXperiments and Technologies. Association for Computing Machinery, New York, NY, USA, 154–167

    work page 2021

  26. [26]

    Jinwoo Park, Byungkwon Choi, Chunghan Lee, and Dongsu Han. 2024. Graph neural network-based SLO-aware proactive resource autoscaling framework for microservices.IEEE/ACM Transactions on Networking32, 4 (2024), 3331–3346

    work page 2024

  27. [27]

    Luan Pham, Huong Ha, and Hongyu Zhang. 2024. Baro: Robust root cause analysis for microservices via multivariate bayesian online change point detection. Proceedings of the ACM on Software Engineering1, FSE (2024), 2214–2237

    work page 2024

  28. [28]

    Luan Pham, Hongyu Zhang, Huong Ha, Flora Salim, and Xiuzhen Zhang. 2025. RCAEval: A Benchmark for Root Cause Analysis of Microservice Systems with Telemetry Data. InCompanion Proceedings of the ACM on Web Conference 2025. Association for Computing Machinery, New York, NY, USA, 777–780

    work page 2025

  29. [29]

    Haoran Qiu, Subho S Banerjee, Saurabh Jha, Zbigniew T Kalbarczyk, and Ravis- hankar K Iyer. 2020. {FIRM}: An intelligent fine-grained resource management framework for {SLO-Oriented} microservices. In14th USENIX symposium on operating systems design and implementation (OSDI 20). USENIX Association, Berkeley, CA, USA, 805–825

    work page 2020

  30. [30]

    Haoran Qiu, Weichao Mao, Chen Wang, Hubertus Franke, Alaa Youssef, Zbig- niew T Kalbarczyk, Tamer Başar, and Ravishankar K Iyer. 2023. {AWARE}: Automate workload autoscaling with reinforcement learning in production cloud systems. In2023 USENIX Annual Technical Conference (USENIX ATC 23). USENIX Association, Berkeley, CA, USA, 387–402

    work page 2023

  31. [31]

    Matías Richart, Juan-Luis Gorricho, Javier Baliosian, Luis M Contreras, Alejandro Muñiz, and Joan Serrat. 2025. LQ-GNN: a Graph Neural Network Model for Response Time Prediction of Microservice-based Applications in the Computing Continuum.IEEE Transactions on Parallel and Distributed Systems36, 12 (2025), 2566–2577

    work page 2025

  32. [32]

    Krzysztof Rzadca, Pawel Findeisen, Jacek Swiderski, Przemyslaw Zych, Przemys- law Broniek, Jarek Kusmierek, Pawel Nowak, Beata Strack, Piotr Witusowski, Steven Hand, et al. 2020. Autopilot: workload autoscaling at google. InProceed- ings of the Fifteenth European Conference on Computer Systems. Association for Computing Machinery, New York, NY, USA, 1–16

    work page 2020

  33. [33]

    José Santos, Efstratios Reppas, Tim Wauters, Bruno Volckaert, and Filip De Turck

    work page

  34. [34]

    Gwydion: Efficient auto-scaling for complex containerized applications in Kubernetes through Reinforcement Learning.Journal of Network and Computer Applications234 (2025), 104067

    work page 2025

  35. [35]

    Jiuchen Shi, Hang Zhang, Zhixin Tong, Quan Chen, Kaihua Fu, and Minyi Guo

    work page

  36. [36]

    In2023 USENIX Annual Technical Con- ference (USENIX ATC 23)

    Nodens: Enabling resource efficient and fast {QoS} recovery of dynamic microservice applications in datacenters. In2023 USENIX Annual Technical Con- ference (USENIX ATC 23). USENIX Association, Berkeley, CA, USA, 403–417

    work page

  37. [37]

    Da Sun Handason Tam, Yang Liu, Huanle Xu, Siyue Xie, and Wing Cheong Lau. 2023. Pert-gnn: Latency prediction for microservice-based cloud-native applications via graph neural networks. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, New York, NY, USA, 2155–2165

    work page 2023

  38. [38]

    Lei Tao, Xianglin Lu, Shenglin Zhang, Jiaqi Luan, Yingke Li, Mingjie Li, Zeyan Li, Qingyang Yu, Hucheng Xie, Ruijie Xu, et al. 2024. Diagnosing performance issues for large-scale microservice systems with heterogeneous graph.IEEE Transactions on Services Computing17, 5 (2024), 2223–2235

    work page 2024

  39. [39]

    The Istio Authors. 2024. Istio: An Open Platform to Connect, Manage, and Secure Microservices. https://istio.io/

    work page 2024

  40. [40]

    The Jaeger Authors. 2018. Jaeger: End-to-End Distributed Tracing. https://www. jaegertracing.io/

    work page 2018

  41. [41]

    2024.Locust: An Open Source Load Testing Tool

    The Locust Authors. 2024.Locust: An Open Source Load Testing Tool. Locust Authors. https://locust.io

    work page 2024

  42. [42]

    Zibo Wang, Pinghe Li, Chieh-Jan Mike Liang, Feng Wu, and Francis Y Yan. 2024. Autothrottle: A Practical {Bi-Level} Approach to Resource Management for {SLO-Targeted} Microservices. In21st USENIX Symposium on Networked Systems Design and Implementation (NSDI 24). USENIX Association, Berkeley, CA, USA, 149–165

    work page 2024

  43. [43]

    Ziliang Wang, Shiyi Zhu, Jianguo Li, Wei Jiang, KK Ramakrishnan, Meng Yan, Xiaohong Zhang, and Alex X Liu. 2024. Deepscaling: Autoscaling microservices with stable cpu utilization for large scale production cloud systems.IEEE/ACM Transactions on Networking32, 5 (2024), 3961–3976

    work page 2024

  44. [44]

    Haixu Wu, Tengge Hu, Yong Liu, Hang Zhou, Jianmin Wang, and Mingsheng Long. 2023. Timesnet: Temporal 2d-variation modeling for general time series analysis. In International Conference on Learning Representations (ICLR)

    work page 2023

  45. [45]

    Qitian Wu, Wentao Zhao, Chenxiao Yang, Hengrui Zhang, Fan Nie, Haitian Jiang, Yatao Bian, and Junchi Yan. 2023. Sgformer: Simplifying and empowering transformers for large-graph representations.Advances in Neural Information Processing Systems36 (2023), 64753–64773

    work page 2023

  46. [46]

    Shuaiyu Xie, Jian Wang, Bing Li, Zekun Zhang, Duantengchuan Li, and Patrick CK Hung. 2024. PBScaler: A bottleneck-aware autoscaling framework for microservice-based applications.IEEE Transactions on Services Computing17, 2 (2024), 604–616

    work page 2024

  47. [47]

    Ruyue Xin, Peng Chen, and Zhiming Zhao. 2023. Causalrca: Causal inference based precise fine-grained root cause localization for microservice applications. Journal of Systems and Software203 (2023), 111724

    work page 2023

  48. [48]

    Huanle Xu, Yang Liu, Siyue Xie, Wing Cheong Lau, et al. 2025. FastPERT: To- wards Fast Microservice Application Latency Prediction via Structural Inductive Bias over PERT Networks. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. AAAI Press, Palo Alto, CA, USA, 20787–20795

    work page 2025

  49. [49]

    Shuangfei Zhai, Walter Talbott, Nitish Srivastava, Chen Huang, Hanlin Goh, Ruixiang Zhang, and Josh Susskind. 2021. An attention free transformer. arXiv preprint arXiv:2105.14103

    work page arXiv 2021

  50. [50]

    Yanqi Zhang, Weizhe Hua, Zhuangzhuang Zhou, G Edward Suh, and Christina Delimitrou. 2021. Sinan: ML-based and QoS-aware resource management for cloud microservices. InProceedings of the 26th ACM international conference on ar- chitectural support for programming languages and operating systems. Association for Computing Machinery, New York, NY, USA, 167–181

    work page 2021

  51. [51]

    Yanqi Zhang, Zhuangzhuang Zhou, Sameh Elnikety, and Christina Delimitrou

    work page

  52. [52]

    In 2024 IEEE International Symposium on High-Performance Computer Architecture (HPCA)

    Ursa: Lightweight resource management for cloud-native microservices. In 2024 IEEE International Symposium on High-Performance Computer Architecture (HPCA). IEEE, IEEE, Piscataway, NJ, USA, 954–969

    work page 2024

  53. [53]

    Laiping Zhao, Yanan Yang, Kaixuan Zhang, Xiaobo Zhou, Tie Qiu, Keqiu Li, and Yungang Bao. 2020. Rhythm: component-distinguishable workload deployment in datacenters. InProceedings of the Fifteenth European Conference on Computer Systems. Association for Computing Machinery, New York, NY, USA, 1–17

    work page 2020

  54. [54]

    Zhiqiang Zhou, Chaoli Zhang, Lingna Ma, Jing Gu, Huajie Qian, Qingsong Wen, Liang Sun, Peng Li, and Zhimin Tang. 2023. AHPA: adaptive horizontal pod autoscaling systems on alibaba cloud container service for kubernetes. InPro- ceedings of the AAAI conference on artificial intelligence, Vol. 37. AAAI Press, Palo Alto, CA, USA, 15621–15629

    work page 2023