Pith. sign in

REVIEW 1 cited by

Learning from the Past: Adaptive Parallelism Tuning for Stream Processing Systems

T0 review · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read StreamTune learns from historical streaming-job DAGs to recommend operator parallelism by predicting operator-level bottlenecks with a monotonic constraint.

arxiv 2504.12074 v2 pith:M6NGARUM submitted 2025-04-16 cs.DC cs.DB

Yuxing Han , Lixiang Chen , Haoyu Wang , Zhanghao Chen , Yifan Zhang , Chengcheng Yang , Kongzhang Hao , Zhengyi Yang This is my paper · ORCID
classification cs.DCcs.DB
keywords parallelismprocessingstreamtunestreamtuningperformancesystemsadaptive
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Stream processing systems like Apache Flink run continuous jobs structured as a graph of operators. Each operator can be run with multiple parallel copies. Choosing too few copies causes backpressure and latency; too many wastes CPU. Today, system operators either use rules or machine-learning models to adjust parallelism. StreamTune is a two-stage method. First, it takes execution histories of many jobs, clusters similar job graphs by how many edits it takes to transform one graph into another, and trains a graph neural network per cluster to predict which operator in a graph would become a bottleneck. Second, when a new job arrives, StreamTune picks the nearest cluster, fine-tunes a lightweight classifier with a monotonic rule: more parallelism never makes an operator more likely to be a bottleneck. It then recommends, for each operator, the smallest parallelism degree the classifier says will avoid a bottleneck, deploys those settings, watches for backpressure, and repeats with the new feedback.

The evaluation was run on Apache Flink and Timely Dataflow using the Nexmark benchmark and PQP synthetic queries. StreamTune reports lower total parallelism than DS2, ContTune, and ZeroTune in most experiments, with zero backpressure occurrences in the tested runs, and latency distributions comparable to the baselines in Timely Dataflow. The largest headline reduction is 83.3% less parallelism than DS2 on Nexmark Q8.

The main caveat is that the pre-training data appears to include the same queries that are later tuned, so the reported gains may partly come from seeing those jobs before. Only one held-out 2-way join job is tested for adaptation time, not for the headline resource savings. No code or data is provided.

Extended reading notes

Core claim

StreamTune reduces reconfigurations by up to 29.6% and parallelism degrees by up to 30.8% in Apache Flink, and achieves up to an 83.3% reduction in parallelism degrees while maintaining comparable processing performance in Timely Dataflow, when compared to state-of-the-art methods. This is the load-bearing empirical assertion: the method is better than DS2, ContTune, and ZeroTune at finding resource-efficient parallelism without causing backpressure.

Load-bearing premise

The monotonic constraint: for every operator and every workload, increasing an operator's parallelism monotonically decreases the probability that it becomes a bottleneck, so a binary search can find the minimal safe parallelism. Section IV-B (Monotonic Constraint of Mf) grounds this in one Flink job with a filter and a window operator, shown in Fig. 4. If parallelism can hurt performance through partitioning, state shuffling, or communication overhead, the recommended parallelism may be wrong.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method rests on several hand-set thresholds and an untested monotonicity assumption; these are the cost of the bottleneck-labeling scheme. No invented physical entities are introduced.

free parameters (5)
  • Bottleneck CPU threshold T = unspecified (e.g., 60%)
    Algorithm 1 labels a downstream operator as bottleneck when resource utilization exceeds T; T changes the training labels and therefore every recommendation.
  • Flink backpressure ratio = 10% of cumulative busy/idle/backpressured time
    Section V-B: a Flink operator is a bottleneck if backPressuredTimeMsPerSecond exceeds 10% of the sum of the three time metrics over an interval; this defines the supervised signal.
  • Timely input-rate ratio = 85%
    Section V-B: a Timely operator is a bottleneck if its input rate falls below 85% of combined upstream output rates; again this defines labels.
  • GED similarity-search threshold tau = 5
    Section V-A sets tau = 5 when computing similarity centers via AStar+-LSa; tau controls which DAGs count as similar in clustering.
  • Cluster count k = determined by elbow method
    Section V-A: k is chosen using the elbow method on the historical DAG set; the number of clusters affects the encoder each target job inherits.
assumptions (4)
  • domain assumption Bottleneck probability decreases monotonically with operator parallelism.
    Section IV-B, Monotonic Constraint of Mf. It is validated on one Flink job with filter and window operators (Fig. 4), then assumed for all operators and workloads.
  • domain assumption Backpressure propagates upstream and bottleneck operators can be identified by checking only the downstream operators of backpressured operators.
    Algorithm 1 labels bottlenecks based on downstream operators' CPU utilization when upstream operators are under backpressure; operators further upstream are left unlabeled because their input rates are distorted.
  • standard math Graph Edit Distance with the two added edit operations satisfies the triangle inequality and is a valid clustering distance.
    Section IV-C uses GED as a metric for K-means; the paper does not prove that the custom operator-type and edge-direction operations preserve the metric property.
  • ad hoc to paper The similarity center, the most frequent DAG within threshold, approximates the true cluster centroid.
    Definition 2 in Section IV-C defines the similarity center; the paper argues but does not validate that it is a good proxy for the median graph.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning from the Past: Adaptive Parallelism Tuning for Stream Processing Systems." pith.science (2026). https://pith.science/paper/M6NGARUM

@misc{pith2026250412074,
  author       = {Pith},
  title        = {Pith review of: Learning from the Past: Adaptive Parallelism Tuning for Stream Processing Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M6NGARUM}},
  note         = {Machine review of arXiv:2504.12074}
}
read the original abstract

Distributed stream processing systems rely on the dataflow model to define and execute streaming jobs, organizing computations as Directed Acyclic Graphs (DAGs) of operators. Adjusting the parallelism of these operators is crucial to handling fluctuating workloads efficiently while balancing resource usage and processing performance. However, existing methods often fail to effectively utilize execution histories or fully exploit DAG structures, limiting their ability to identity bottlenecks and determine the optimal parallelism. In this paper, we propose StreamTune, a novel approach for adaptive paralelism tuning in stream processing systems. StreamTune incorporates a pre-training and fine-tuning framework that leverages global knowledge from historical execution data for job-specific parallelism tuning. In the pre-training phase, Stream Tune clusters the historical data with Graph Edit Distance and pre-trains a Graph Neural Networkbased encoder per cluster to capture the correlation between the operator parallelism, DAG structures, and the identified operator-level bottlenecks. In the online tuning phase, StreamTune iteratively refines operator parallelism recommendations using an operator-level bottleneck prediction model enforced with a monotonic constraint, which aligns with the observed system performance behavior. Evaluation results demonstrate that StreamTune reduces reconfigurations by up to 29.6% and parallelism degrees by up to 30.8% in Apache Flink under a synthetic workload. In Timely Dataflow, StreamTune achieves up to an 83.3% reduction in parallelism degrees while maintaining comparable processing performance under the Nexmark benchmark, when compared to the state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2504.12074 by the authors.

Figure 1
Figure 1. A Logic Dataflow DAG. efficiency. Dataflow-level backpressure reflects a system-wide issue caused by one or more bottlenecks, while operator￾level backpressure identifies the specific operators within the dataflow DAG responsible for the performance degradation. Dataflow Execution Histories. A long-running stream pro￾cessing system accumulates extensive execution histories from various streaming jobs, providing valu… view at source ↗
Figure 2
Figure 2. Two-Phase Parallelism Tuning. bottleneck prediction. The streaming job is redeployed with the recommended parallelisms, and its runtime feedback is collected to refine the prediction layer. This iterative pro￾cess continues until the bottleneck is fully mitigated for the streaming job or the recommended parallelism degrees remain unchanged across iterations. A monotonic constraint is applied to the bottleneck pre￾di… view at source ↗
Figure 3
Figure 3. provides an example of bottleneck identification in a dataflow DAG. In this DAG, operator O1 is under backpres￾sure due to the limited processing ability of its downstream operators, O2 and O3. The CPU utilization of O3 is measured at 15%, whereas O2 exhibits a significantly higher CPU load of 98%, exceeding the predefined threshold. As a result, O2 is Algorithm 1: Bottleneck Identification. Input: A dataflow DAG g … view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Relationship Between Parallelism and Processing Ability. operators as Olabel. The total loss for the labeled set is computed as the average loss across all labeled operators: Ltotal = 1 |Olabel| P v∈Olabel Lv. B. Online Fine-tuning We begin by presenting the workflow o…
Figure 5
Figure 5. Figure 5: Distribution of Pre-trained Dataflow DAGs rate over time for the evaluated workloads using a periodic pattern. Specifically, a basic cycle of ten source rates can be defined as [3Wu, 7Wu, 4Wu, 2Wu, 1Wu, 10Wu, 8Wu, 5Wu, 6Wu, 9Wu], where Wu denotes the source rate unit, …
Figure 6
Figure 6. Figure 6: Final parallelism recommendations by different methods when the source data rate is [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Reconfiguration efficiency and adaptation time of StreamTune TABLE III: Frequency of Backpressure Occurrences Query Method Q1 Q2 Q3 Q5 Q8 Linear 2-way-join 3-way-join DS2 0 0 1 2 1 3 8 12 ContTune 0 0 2 5 1 4 11 9 ZeroTune / / / / / 0 0 0 StreamTune 0 0 0 0 0 0 0 0 ope…
Figure 8
Figure 8. Figure 8: Evaluation Results in Timely Dataflow. F. Generality Evaluation on Timely DataFlow To demonstrate the generality of StreamTune, we conduct a comparative evaluation on Timely Dataflow. We report results for Q3, Q5, and Q8, as other Nexmark jobs run effectively with a pa…
Figure 11
Figure 11. Figure 11: Ablation Study. GED computation grows significantly. In contrast, AStar+- LSa maintains consistently low processing times. Notably, for a dataset with 400 dataflow DAGs, it reduces the time cost by 99.65% compared to direct GED computation. VI. RELATED WORK Parallelis…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Towards a Proactive Autoscaling Framework for Data Stream Processing at the Edge using GRU and Transfer Learning

    cs.DC 2025-07 reject novelty 3.0 of 10

    A GRU forecaster beats CNN, ARIMA, and Prophet on simulated and rescaled load data for edge stream processing, but the transfer-learning and autoscaling stages that complete the proposed framework are not implemented.

Reference graph

Works this paper leans on

60 extracted references · 49 canonical work pages · cited by 1 Pith paper

  1. [1]

    Aapche Flink,

    “Aapche Flink,” https://flink.apache.org/, [Accessed 2024/10/28]

    work page 2024

  2. [2]

    A Modular Implementation of Timely Dataflow in Rust,

    F. McSherry, “A Modular Implementation of Timely Dataflow in Rust,” https://github.com/frankmcsherry/timely-dataflow/, [Accessed 2024/10/28]

    work page 2024

  3. [3]

    Aapche Samza,

    “Aapche Samza,” https://samza.apache.org/, [Accessed 2024/10/28]

    work page 2024

  4. [4]

    Amazon Kinesis,

    “Amazon Kinesis,” https://aws.amazon.com/kinesis/, [Accessed 2024/10/28]

    work page 2024

  5. [5]

    Aapche Hadoop,

    “Aapche Hadoop,” https://hadoop.apache.org/, [Accessed 2024/10/28]

    work page 2024

  6. [6]

    Aapche Pig,

    “Aapche Pig,” https://hadoop.apache.org/, [Accessed 2024/10/28]

    work page 2024

  7. [7]

    Aapche Hive,

    “Aapche Hive,” https://hive.apache.org/, [Accessed 2024/10/28]

    work page 2024

  8. [8]

    StreamOps: Cloud-native runtime management for streaming services in bytedance,

    Y . Mao, Z. Chen, Y . Zhang, M. Wang, Y . Fang, G. Zhang, R. Shi, and R. T. Ma, “StreamOps: Cloud-native runtime management for streaming services in bytedance,” Proceedings of the VLDB Endowment , vol. 16, no. 12, pp. 3501–3514, 2023

    work page 2023

Show all 60 references
  1. [9]

    The dataflow model: a practical approach to balancing correctness, latency, and cost in massive-scale, unbounded, out-of-order data pro- cessing,

    T. Akidau, R. Bradshaw, C. Chambers, S. Chernyak, R. J. Fern ´andez- Moctezuma, R. Lax, S. McVeety, D. Mills, F. Perry, E. Schmidt et al., “The dataflow model: a practical approach to balancing correctness, latency, and cost in massive-scale, unbounded, out-of-order data pro- ...

    2015

  2. [10]

    Naiad: a timely dataflow system,

    D. G. Murray, F. McSherry, R. Isaacs, M. Isard, P. Barham, and M. Abadi, “Naiad: a timely dataflow system,” in Proceedings of the Twenty-Fourth ACM Symposium on Operating Systems Principles, 2013, pp. 439–455

    2013

  3. [11]

    Lightweight asynchronous snapshots for distributed dataflows,

    P. Carbone, G. F ´ora, S. Ewen, S. Haridi, and K. Tzoumas, “Lightweight asynchronous snapshots for distributed dataflows,” arXiv preprint arXiv:1506.08603, 2015

    2015 arXiv

  4. [12]

    Turbine: Facebook’s service management platform for stream processing,

    Y . Mei, L. Cheng, V . Talwar, M. Y . Levin, G. Jacques-Silva, N. Simha, A. Banerjee, B. Smith, T. Williamson, S. Yilmaz et al. , “Turbine: Facebook’s service management platform for stream processing,” in ICDE, 2020, pp. 1591–1602

    2020

  5. [13]

    Dhalion: self-regulating stream processing in heron,

    A. Floratou, A. Agrawal, B. Graham, S. Rao, and K. Ramasamy, “Dhalion: self-regulating stream processing in heron,” Proceedings of the VLDB Endowment , vol. 10, no. 12, pp. 1825–1836, 2017

    2017

  6. [14]

    Three steps is all you need: fast, accurate, automatic scaling decisions for distributed streaming dataflows,

    V . Kalavri, J. Liagouris, M. Hoffmann, D. Dimitrova, M. Forshaw, and T. Roscoe, “Three steps is all you need: fast, accurate, automatic scaling decisions for distributed streaming dataflows,” in OSDI, 2018, pp. 783– 798

    2018

  7. [15]

    DRS: Auto-scaling for real-time stream analytics,

    T. Z. Fu, J. Ding, R. T. Ma, M. Winslett, Y . Yang, and Z. Zhang, “DRS: Auto-scaling for real-time stream analytics,” IEEE/ACM Transactions on networking, vol. 25, no. 6, pp. 3338–3352, 2017

    2017

  8. [16]

    Elastic stream processing with latency guarantees,

    B. Lohrmann, P. Janacik, and O. Kao, “Elastic stream processing with latency guarantees,” in 2015 IEEE 35th International Conference on Distributed Computing Systems . IEEE, 2015, pp. 399–410

    2015

  9. [17]

    Integrating scale out and fault tolerance in stream processing using operator state management,

    R. Castro Fernandez, M. Migliavacca, E. Kalyvianaki, and P. Pietzuch, “Integrating scale out and fault tolerance in stream processing using operator state management,” in SIGMOD, 2013, pp. 725–736

    2013

  10. [18]

    Stela: Enabling stream processing systems to scale-in and scale-out on-demand,

    L. Xu, B. Peng, and I. Gupta, “Stela: Enabling stream processing systems to scale-in and scale-out on-demand,” in 2016 IEEE International Conference on Cloud Engineering (IC2E) . IEEE, 2016, pp. 22–31

    2016

  11. [19]

    ContTune: Continuous tuning by conservative bayesian optimization for distributed stream data processing systems,

    J. Lian, X. Zhang, Y . Shao, Z. Pu, Q. Xiang, Y . Li, and B. Cui, “ContTune: Continuous tuning by conservative bayesian optimization for distributed stream data processing systems,” Proceedings of the VLDB Endowment, vol. 16, no. 13, pp. 4282–4295, 2023

    2023

  12. [20]

    ZeroTune: Learned zero-shot cost models for parallelism tuning in stream processing,

    P. Agnihotri, B. Koldehofe, P. Stiegele, R. Heinrich, C. Binnig, and M. Luthra, “ZeroTune: Learned zero-shot cost models for parallelism tuning in stream processing,” in ICDE, 2024, pp. 2040–2053

    2024

  13. [21]

    A review of generalized zero-shot learning meth- ods,

    F. Pourpanah, M. Abdar, Y . Luo, X. Zhou, R. Wang, C. P. Lim, X.-Z. Wang, and Q. J. Wu, “A review of generalized zero-shot learning meth- ods,” IEEE transactions on pattern analysis and machine intelligence , vol. 45, no. 4, pp. 4051–4070, 2022

    2022

  14. [22]

    CSI GED: An efficient approach for graph edit similarity computation,

    K. Gouda and M. Hassaan, “CSI GED: An efficient approach for graph edit similarity computation,” in 2016 IEEE 32nd International Conference on Data Engineering (ICDE) . IEEE, 2016, pp. 265–276

    2016

  15. [23]

    Efficient graph similarity search over large graph databases,

    W. Zheng, L. Zou, X. Lian, D. Wang, and D. Zhao, “Efficient graph similarity search over large graph databases,” IEEE Transactions on Knowledge and Data Engineering , vol. 27, no. 4, pp. 964–978, 2014

    2014

  16. [24]

    A partition-based approach to structure similarity search,

    X. Zhao, C. Xiao, X. Lin, Q. Liu, and W. Zhang, “A partition-based approach to structure similarity search,” PVLDB, vol. 7, no. 3, pp. 169– 180, 2013

    2013

  17. [25]

    How powerful are graph neural networks?

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful are graph neural networks?” in ICLR 2019,. OpenReview.net, 2019

    2019

  18. [26]

    Inductive representation learning on large graphs,

    W. L. Hamilton, Z. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” in NIPS 2017 , I. Guyon, U. von Luxburg, S. Bengio, H. M. Wallach, R. Fergus, S. V . N. Vishwanathan, and R. Garnett, Eds., 2017, pp. 1024–1034

    2017

  19. [27]

    Representation learning on graphs with jumping knowledge networks,

    K. Xu, C. Li, Y . Tian, T. Sonobe, K.-i. Kawarabayashi, and S. Jegelka, “Representation learning on graphs with jumping knowledge networks,” in ICML, 2018, pp. 5453–5462

    2018

  20. [28]

    Neural message passing for quantum chemistry,

    J. Gilmer, S. S. Schoenholz, P. F. Riley, O. Vinyals, and G. E. Dahl, “Neural message passing for quantum chemistry,” in ICML. PMLR, 2017, pp. 1263–1272

    2017

  21. [29]

    Meta-gnn: On few-shot node classification in graph meta-learning,

    F. Zhou, C. Cao, K. Zhang, G. Trajcevski, T. Zhong, and J. Geng, “Meta-gnn: On few-shot node classification in graph meta-learning,” in Proceedings of the 28th ACM International Conference on Information and Knowledge Management , 2019, pp. 2357–2360

    2019

  22. [30]

    N-gcn: Multi- scale graph convolution for semi-supervised node classification,

    S. Abu-El-Haija, A. Kapoor, B. Perozzi, and J. Lee, “N-gcn: Multi- scale graph convolution for semi-supervised node classification,” in uncertainty in artificial intelligence . PMLR, 2020, pp. 841–851

    2020

  23. [31]

    Evolvegcn: Evolving graph convolutional networks for dynamic graphs,

    A. Pareja, G. Domeniconi, J. Chen, T. Ma, T. Suzumura, H. Kanezashi, T. Kaler, T. Schardl, and C. Leiserson, “Evolvegcn: Evolving graph convolutional networks for dynamic graphs,” in Proceedings of the AAAI conference on artificial intelligence , vol. 34, no. 04, 2020, pp. 5363– 5370

    2020

  24. [32]

    Link prediction based on graph neural net- works,

    M. Zhang and Y . Chen, “Link prediction based on graph neural net- works,” Advances in neural information processing systems , vol. 31, 2018

    2018

  25. [33]

    Learning to rep- resent programs with graphs,

    M. Allamanis, M. Brockschmidt, and M. Khademi, “Learning to rep- resent programs with graphs,” in International Conference on Learning Representations, 2018

    2018

  26. [34]

    Text level graph neural network for text classification,

    L. Huang, D. Ma, S. Li, X. Zhang, and H. Wang, “Text level graph neural network for text classification,” in Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-...

    2019

  27. [35]

    One model to rule them all: Towards zero- shot learning for databases,

    B. Hilprecht and C. Binnig, “One model to rule them all: Towards zero- shot learning for databases,” in CIDR, 2022

    2022

  28. [36]

    A unified transferable model for ml-enhanced dbms,

    Z. Wu, P. Yang, P. Yu, R. Zhu, Y . Han, Y . Li, D. Lian, K. Zeng, and J. Zhou, “A unified transferable model for ml-enhanced dbms,” CIDR, 2022

    2022

  29. [37]

    LlamaTune: sample-efficient dbms configuration tuning,

    K. Kanellis, C. Ding, B. Kroth, A. M ¨uller, C. Curino, and S. Venkatara- man, “LlamaTune: sample-efficient dbms configuration tuning,” vol. 15, no. 11, p. 2953–2965, 2022

    2022

  30. [38]

    The cross entropy method for classification,

    S. Mannor, D. Peleg, and R. Rubinstein, “The cross entropy method for classification,” in Proceedings of the 22nd international conference on Machine learning, 2005, pp. 561–568

    2005

  31. [39]

    Certified monotonic neural networks,

    X. Liu, X. Han, N. Zhang, and Q. Liu, “Certified monotonic neural networks,” in NIPS, 2020

    2020

  32. [40]

    Constrained monotonic neural networks,

    D. Runje and S. M. Shankaranarayana, “Constrained monotonic neural networks,” in ICML, A. Krause, E. Brunskill, K. Cho, B. Engelhardt, S. Sabato, and J. Scarlett, Eds., vol. 202. PMLR, 2023, pp. 29 338– 29 353

    2023

  33. [41]

    Support-vector networks,

    C. Cortes, “Support-vector networks,” Machine Learning, 1995

    1995

  34. [42]

    A training algorithm for optimal margin classifiers,

    B. E. Boser, I. M. Guyon, and V . N. Vapnik, “A training algorithm for optimal margin classifiers,” in Proceedings of the fifth annual workshop on Computational learning theory , 1992, pp. 144–152

    1992

  35. [43]

    Xgboost: A scalable tree boosting system,

    T. Chen and C. Guestrin, “Xgboost: A scalable tree boosting system,” in Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , 2016, pp. 785–794

    2016

  36. [44]

    Lightgbm: A highly efficient gradient boosting decision tree,

    G. Ke, Q. Meng, T. Finley, T. Wang, W. Chen, W. Ma, Q. Ye, and T.- Y . Liu, “Lightgbm: A highly efficient gradient boosting decision tree,” NIPS, vol. 30, 2017

    2017

  37. [45]

    A distance measure between attributed relational graphs for pattern recognition,

    A. Sanfeliu and K.-S. Fu, “A distance measure between attributed relational graphs for pattern recognition,” IEEE transactions on systems, man, and cybernetics , no. 3, pp. 353–362, 1983

    1983

  38. [46]

    Boosting graph similarity search through pre-computation,

    J. Kim, “Boosting graph similarity search through pre-computation,” in SIGMOD, 2021, pp. 951–963

    2021

  39. [47]

    Comparing stars: On approximating graph edit distance,

    Z. Zeng, A. K. Tung, J. Wang, J. Feng, and L. Zhou, “Comparing stars: On approximating graph edit distance,” VLDB, vol. 2, no. 1, pp. 25–36, 2009

    2009

  40. [48]

    Computing similarity between rna structures,

    B. Ma, L. Wang, and K. Zhang, “Computing similarity between rna structures,” Theoretical Computer Science , vol. 276, no. 1-2, pp. 111– 132, 2002

    2002

  41. [49]

    The k-means algorithm: A comprehensive survey and performance evaluation,

    M. Ahmed, R. Seraj, and S. M. S. Islam, “The k-means algorithm: A comprehensive survey and performance evaluation,” Electronics, vol. 9, no. 8, p. 1295, 2020

    2020

  42. [50]

    An median graphs: properties, algorithms, and applications,

    X. Jiang, A. Munger, and H. Bunke, “An median graphs: properties, algorithms, and applications,” IEEE Transactions on pattern analysis and machine intelligence , vol. 23, no. 10, pp. 1144–1151, 2001

    2001

  43. [51]

    Speeding up GED verification for graph similarity search,

    L. Chang, X. Feng, X. Lin, L. Qin, W. Zhang, and D. Ouyang, “Speeding up GED verification for graph similarity search,” in 2020 IEEE 36th International Conference on Data Engineering (ICDE) . IEEE, 2020, pp. 793–804

    2020

  44. [52]

    Inves: Incremental partitioning-based verification for graph similarity search

    J. Kim, D.-H. Choi, and C. Li, “Inves: Incremental partitioning-based verification for graph similarity search.” in EDBT, 2019, pp. 229–240

    2019

  45. [53]

    An exact graph edit distance algorithm for solving pattern recognition problems,

    Z. Abu-Aisheh, R. Raveaux, J.-Y . Ramel, and P. Martineau, “An exact graph edit distance algorithm for solving pattern recognition problems,” in 4th International Conference on Pattern Recognition Applications and Methods 2015, 2015

    2015

  46. [54]

    Nexmark benchmark,

    “Nexmark benchmark,” https://beam.apache.org/documentation/sdks/ java/testing/nexmark/, 2024, [Accessed 2024/10/28]

    2024

  47. [55]

    The application of cluster analysis in strategic management research: an analysis and critique,

    D. J. Ketchen and C. L. Shook, “The application of cluster analysis in strategic management research: an analysis and critique,” Strategic management journal, vol. 17, no. 6, pp. 441–458, 1996

    1996

  48. [56]

    DS2 github repository,

    “DS2 github repository,” https://github.com/strymon-system/ds2, 2024, [Accessed 2024/10/28]

    2024

  49. [57]

    Gml: effi- ciently auto-tuning flink’s configurations via guided machine learning,

    Y . Guo, H. Shan, S. Huang, K. Hwang, J. Fan, and Z. Yu, “Gml: effi- ciently auto-tuning flink’s configurations via guided machine learning,” IEEE Transactions on Parallel and Distributed Systems, vol. 32, no. 12, pp. 2921–2935, 2021

    2021

  50. [58]

    Automatic database management system tuning through large-scale machine learn- ing,

    D. Van Aken, A. Pavlo, G. J. Gordon, and B. Zhang, “Automatic database management system tuning through large-scale machine learn- ing,” in Proceedings of the 2017 ACM international conference on management of data , 2017, pp. 1009–1024

    2017

  51. [59]

    Restune: Resource oriented tuning boosted by meta-learning for cloud databases,

    X. Zhang, H. Wu, Z. Chang, S. Jin, J. Tan, F. Li, T. Zhang, and B. Cui, “Restune: Resource oriented tuning boosted by meta-learning for cloud databases,” in SIGMOD, 2021, pp. 2102–2114

    2021

  52. [60]

    An end-to-end automatic cloud database tuning system using deep reinforcement learning,

    J. Zhang, Y . Liu, K. Zhou, G. Li, Z. Xiao, B. Cheng, J. Xing, Y . Wang, T. Cheng, L. Liu et al., “An end-to-end automatic cloud database tuning system using deep reinforcement learning,” in SIGMOD, 2019, pp. 415– 432

    2019

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.