Horizontal vs. Vertical Executor Scaling for TPC-DS Analytical Workloads on Geo-Distributed Kubernetes

• Italo V. P. Guimaraes UnB https://orcid.org/0009-0000-2291-1807
• Aleteia Araujo UnB https://orcid.org/0000-0003-4645-6700

DOI: https://doi.org/10.5753/semish.2026.23385

Resumo

When deploying Apache Spark on Kubernetes, practitioners must decide whether to allocate resources as many small executors or as few large ones. The conventional argument for large executors that they reduce shuffle redistribution rests on an assumption that Adaptive Query Execution (AQE) in Spark 3.5 renders false: shuffle volume is virtually identical across all executor sizes, because AQE adapts partition plans to the data at runtime, not to executor configuration. Without a shuffle advantage, vertical scaling only adds JVM garbage-collection overhead: large executors accumulate up to 84% more GC time, and horizontal scaling delivers 22% shorter execution at the same resource budget. Practitioners should prefer small executors (4 cores, 14 GB) and scale out.

Referências

Abe, H. et al. (2023). Evaluating Spark on Kubernetes under different resource configurations. In 2023 IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing. IEEE.

Armbrust, M., Das, T., Torres, L., Yavuz, B., Zhu, S., Xuan, R., Ghodsi, A., Stoica, I., and Zaharia, M. (2020). Delta Lake: High-performance ACID table storage over cloud object stores. Proceedings of the VLDB Endowment, 13(12):3411–3424.

Armbrust, M., Ghodsi, A., Xin, R., and Zaharia, M. (2021). Lakehouse: A new generation of open platforms that unify data warehousing and advanced analytics. In Proceedings of the 11th Conference on Innovative Data Systems Research (CIDR).

Behm, A., Palkar, S., Agarwal, U., Armbrust, M., Cashman, M., Chakraborty, A., Chenghao, A., Das, T., Davidson, A., Dhar, A., Ghodsi, A., Gupta, S., Jaeger, R., Khalapyan, L., Lai, E., Liu, S., Luszczak, A., Sankaranarayanan, H., Soman, B., Stuhrm, T., Sundaram, S., Tsai, A., Wang, Y., Wang, Y., Wu, R., Yavuz, B., Stoica, I., Boncz, P. A., and Zaharia, M. (2022). Photon: A fast query engine for lakehouse systems. In Proceedings of the 2022 ACM SIGMOD International Conference on Management of Data, pages 2326–2339.

Burns, B., Grant, B., Oppenheimer, D., Brewer, E., and Wilkes, J. (2016). Borg, Omega, and Kubernetes. ACM Queue, 14(1):70–93.

Guo, W. et al. (2023). Optimizing analytical workloads on cloud-native Spark. In 2023 IEEE International Conference on Cloud Computing. IEEE.

Huang, G., Chen, Z., et al. (2024). Benchmarking the data lake ecosystem: Delta Lake, Apache Iceberg, and Apache Hudi. In Proceedings of the ACM SIGMOD International Conference on Management of Data.

Li, W. et al. (2022). Performance evaluation of Apache Spark on Kubernetes. In 2022 IEEE International Conference on Big Data, pages 1–8. IEEE.

MinIO, Inc. (2024). MinIO: High performance object storage.

Mondin, L. and Dias, G. (2023). Plataformas flexíveis para experimentação: o caminho para atender as novas demandas de experimentação científica em TICs. In Anais do XIV Workshop de Pesquisa Experimental da Internet do Futuro (WPEIF), pages 25–32, Brasília/DF. Sociedade Brasileira de Computação.

Nambiar, R. O. and Poess, M. (2006). The making of TPC-DS. In Proceedings of the 32nd International Conference on Very Large Data Bases, pages 1049–1058.

Pedrosa, E., Martins, J., Sousa, F., Mondin, L., and Dias, G. (2024). Indo além das simulações com o serviço de testbeds: Ambientes reais para experimentação científica em TICs. In Anais do XV Workshop de Pesquisa Experimental da Internet do Futuro (WPEIF), pages 47–54, Niterói/RJ. Sociedade Brasileira de Computação.

Poess, M., Smith, B., Kollar, L., and Larson, P. (2002). Tpc-ds, taking decision support benchmarking to the next level. In Proceedings of the 2002 ACM SIGMOD International Conference on Management of Data, SIGMOD ’02, page 582–587, New York, NY, USA. Association for Computing Machinery.

Shi, J., Qiu, Y., Minhas, U. F., Jiao, L., Wang, C., Reinwald, B., and Ozcan, F. (2015). Clash of the titans: MapReduce vs. Spark for large scale data analytics. Proceedings of the VLDB Endowment, 8(13):2110–2121.

Tang, Z. et al. (2016). Benchmarking Apache Spark and Hadoop MapReduce on big data workloads. In 2016 IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing. IEEE.

Transaction Processing Performance Council (2021). TPC-DS decision support benchmark, version 3.2.0. Technical report, TPC.

Zaharia, M., Xin, R. S., Wendell, P., Das, T., Armbrust, M., Dave, A., Meng, X., Rosen, J., Venkataraman, S., Franklin, M. J., et al. (2016). Apache Spark: A unified engine for big data processing. Communications of the ACM, 59(11):56–65.

Zhu, C., Han, W., Huang, L., and Ma, Y. (2020). Adaptive query execution: Speeding up Spark SQL at runtime. Proceedings of the VLDB Endowment, 13(12):3533–3546.