Advertisement

Apache Spark Streaming, Kafka and HarmonicIO: A Performance Benchmark and Architecture Comparison for Enterprise and Scientific Computing

Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 12093)

Abstract

Many scientific computing applications generate streams where message sizes exceed one megabyte, in contrast with smaller message sizes in enterprise contexts (order kilobytes, often XML or JSON). Furthermore, the processing cost of messages in scientific computing applications are usually an order of magnitude higher than in typical enterprise applications. Frameworks such as Apache Spark offer high throughput processing of streams with such ‘enterprise’ characteristics, as well as scalability, with high resilience and many other desirable features. Motivated by the development of near real-time image processing pipelines for roboticized microscopy, we evaluate the suitability of Apache Spark for streams more typical of scientific computing applications, those with large message sizes (up to 10 MB), and heavy per-message CPU load, under typical stream integrations. For comparison, we benchmark a P2P stream processing framework, HarmonicIO, developed in-house. Our study reveals a complex interplay of performance trade-offs, revealing the boundaries of good performance for each framework and integration over a wide domain of application loads. Based on these results, we suggest which are likely to offer good performance for a given load. Broadly, the advantages of Spark’s rich features makes its performance sensitive to message size in particular, whereas the simplicity of HarmonicIO offers more robust performance, and better CPU utilization.

Keywords

Stream processing Apache Spark HarmonicIO High-throughput microscopy HPC Benchmark XaaS HASTE 

References

  1. 1.
    Apache Software Foundation: Apache Hadoop. http://hadoop.apache.org/ (2011)
  2. 2.
    Apache Software Foundation: Cassandra Limitations. https://cwiki.apache.org/confluence/display/CASSANDRA2/CassandraLimitations (2019)
  3. 3.
    Ausmees, K., John, A., Toor, S.Z., Hellander, A., Nettelblad, C.: BAMSI: a multi-cloud service for scalable distributed filtering of massive genome data. BMC Bioinform. 19(1), 240 (2018).  https://doi.org/10.1186/s12859-018-2241-zCrossRefGoogle Scholar
  4. 4.
    Blamey, B., Wrede, F., Karlsson, J., Hellander, A., Toor, S.: Adapting the secretary hiring problem for optimal hot-cold tier placement under Top-K workloads. In: 2019 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID). pp. 576–583, May 2019.  https://doi.org/10.1109/CCGRID.2019.00074
  5. 5.
    Carbone, P., Katsifodimos, A., Ewen, S., Markl, V., Haridi, S., Tzoumas, K.: Apache Flink™: stream and batch processing in a single engine. Bull. IEEE Comput. Soc. Techn. Committee Data Eng. 36(4), 28–38 (2015)Google Scholar
  6. 6.
    Chintapalli, S., et al.: Benchmarking streaming computation engines: storm, flink and spark streaming. In: 2016 IEEE International Parallel and Distributed Processing Symposium Workshops, pp. 1789–1792. IEEE (2016)Google Scholar
  7. 7.
    Kreps, J., Narkhede, N., Rao, J., et al.: Kafka: a distributed messaging system for log processing. In: Proceedings of the NetDB, pp. 1–7 (2011)Google Scholar
  8. 8.
    Lugnegård, L.: Building a high throughput microscope simulator using the Apache Kafka streaming framework (M.sc. Thesis) (2018)Google Scholar
  9. 9.
    Marcu, O.C., Costan, A., Antoniu, G., Pérez-Hernández, M.S.: Spark versus flink: understanding performance in big data analytics frameworks. In: 2016 IEEE International Conference on Cluster Computing (CLUSTER), pp. 433–442, September 2016.  https://doi.org/10.1109/CLUSTER.2016.22
  10. 10.
    National Energy Research Scientific Computing Center: Characterization of the DOE Mini-apps. https://portal.nersc.gov/project/CAL/designforward.htm, June 2014
  11. 11.
    Qian, S., Wu, G., Huang, J., Das, T.: Benchmarking modern distributed streaming platforms. In: 2016 IEEE International Conference on Industrial Technology (ICIT), pp. 592–598, March 2016.  https://doi.org/10.1109/ICIT.2016.7474816
  12. 12.
    Salvatore Sanfilippo: Redis (2009). https://redis.io/
  13. 13.
    Toor, S., et al.: SNIC Science Cloud (SSC): a national-scale cloud infrastructure for Swedish academia. In: 2017 IEEE 13th International Conference On E-Science (e-Science), pp. 219–227. IEEE (2017)Google Scholar
  14. 14.
    Torruangwatthana, P., Wieslander, H., Blamey, B., Hellander, A., Toor, S.: HarmonicIO: scalable data stream processing for scientific datasets. In: 2018 IEEE 11th International Conference on Cloud Computing (CLOUD), pp. 879–882, July 2018.  https://doi.org/10.1109/CLOUD.2018.00126
  15. 15.
    Wollman, R., Stuurman, N.: High throughput microscopy: from raw images to discoveries. J. Cell Sci. 120(21), 3715–3722 (2007).  https://doi.org/10.1242/jcs.013623 CrossRefGoogle Scholar
  16. 16.
    Xin, R.: Apache Spark the fastest open source engine for sorting a petabyte, October 2014. https://databricks.com/blog/2014/10/10/spark-petabyte-sort.html
  17. 17.
    Zaharia, M., et al.: Resilient distributed datasets: a fault-tolerant abstraction for in-memory cluster computing. In: Proceedings of the 9th USENIX Conference on Networked Systems Design and Implementation, pp. 2–2. NSDI 2012, USENIX Association, San Jose (2012)Google Scholar
  18. 18.
    Zaharia, M., et al.: Apache spark: a unified engine for big data processing. Commun. ACM 59(11), 56–65 (2016).  https://doi.org/10.1145/2934664CrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2020

Authors and Affiliations

  1. 1.Department of Information Technology, Division of Scientific ComputingUppsala UniversityUppsalaSweden

Personalised recommendations