Communication Models for Resource Constrained Hierarchical Ethernet Networks

  • Jun Zhu
  • Alexey Lastovetsky
  • Shoukat Ali
  • Rolf Riesen
Part of the Lecture Notes in Computer Science book series (LNCS, volume 8374)

Abstract

Communication time prediction is critical for parallel application performance tuning, especially for the rapidly growing field of data-intensive applications. However, making such predictions accurately is non-trivial when contention exists on different components in hierarchical networks. In this paper, we derive an ‘asymmetric network property’ on TCP layer for concurrent bidirectional communications in a commercial off-the-shelf (COTS) cluster, and develop a communication model as the first effort to characterize the communication times on hierarchical Ethernet networks with contentions on both NIC and backbone cable levels. In particular, we show that if the asymmetric network property was excluded from the model, the communication time predictions will be significantly less accurate than those made by using the asymmetric network property. Furthermore, our observation of the performance degradation caused by the asymmetric network property suggests that some part of the software stack below TCP layer in COTS clusters needs targeted tuning, which has not yet attracted any attention in literature.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    Iperf site, http://sourceforge.net/projects/iperf/ (retrieved January 24, 2012)
  2. 2.
    OpenMPI site, http://www.open-mpi.org (retrieved January 24, 2012)
  3. 3.
    Alexandrov, A., Ionescu, M.F., Schauser, K.E., Scheiman, C.: LogGP: incorporating long messages into the LogP model – one step closer towards a realistic model for parallel computation. In: P. of. the Seventh Annual ACM Symposium on Parallel Algorithms and Architectures, SPAA 1995, pp. 95–105 (1995)Google Scholar
  4. 4.
    Bell, G., Gray, J., Szalay, A.: Petascale computational systems. Computer 39, 110–112 (2006)CrossRefGoogle Scholar
  5. 5.
    Leitao, B.H.: Tuning 10Gb network cards on Linux. In: Ottawa Linux Symposium, pp. 169–184 (2009)Google Scholar
  6. 6.
    Culler, D., Karp, R., Patterson, D., Sahay, A., Schauser, K.E., Santos, E., Subramonian, R., von Eicken, T.: LogP: towards a realistic model of parallel computation. SIGPLAN Not. 28, 1–12 (1993)CrossRefGoogle Scholar
  7. 7.
    Gokhale, M., Cohen, J., Yoo, A., Miller, W.M., Jacob, A., Ulmer, C., Pearce, R.: Hardware technologies for high-performance data-intensive computing. Computer 41(4), 60–68 (2008)CrossRefGoogle Scholar
  8. 8.
    Hockney, R.W.: The communication challenge for MPP: Intel Paragon and Meiko CS-2. Parallel Comput. 20, 389–398 (1994)CrossRefGoogle Scholar
  9. 9.
    Hoefler, T., Mehlan, T., Mietke, F., Rehm, W.: LogfP - A Model for small Messages in InfiniBand. In: P. of. the 20th IEEE Int’l Parallel & Distributed Processing Symposium (IPDPS), PMEO-PDS 2006 Workshop (April 2006)Google Scholar
  10. 10.
    Johnston, W.E.: High-speed, wide area, data intensive computing: A ten year retrospective. In: P. of. the 7th IEEE Int’l Symposium on High Performance Distributed Computing, HPDC 1998, pp. 280–291 (1998)Google Scholar
  11. 11.
    Lastovetsky, A., Rychkov, V., O’Flynn, M.: Accurate heterogeneous communication models and a software tool for their efficient estimation. Int’l J. of High Performance Computing Applications 24, 34–48 (2010)CrossRefGoogle Scholar
  12. 12.
    Martinasso, M., Méhaut, J.-F.: A contention-aware performance model for HPC-based networks: A case study of the InfiniBand network. In: Jeannot, E., Namyst, R., Roman, J. (eds.) Euro-Par 2011, Part I. LNCS, vol. 6852, pp. 91–102. Springer, Heidelberg (2011)CrossRefGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2014

Authors and Affiliations

  • Jun Zhu
    • 1
  • Alexey Lastovetsky
    • 2
  • Shoukat Ali
    • 3
  • Rolf Riesen
    • 3
  1. 1.Technical University of EindhovenNetherlands
  2. 2.University College DublinIreland
  3. 3.Dublin Research LaboratoryIBMIreland

Personalised recommendations