Skip to main content
SpringerLink
Log in

Disk cache architectures for transaction-like applications in parallel computers

Platten-Cache-Architekturen für transaktionsartige Anwendungen in Parallelrechnern

  • Published:
Computing Aims and scope Submit manuscript

Abstract

I/O in computer systems is prone to become a bottleneck. This is a particular severe problem in highly parallel machines where some applications are fully I/O bound if only one or few conventional I/O paths exist. Similar to the use of multiprocessor technology for increasing processing performance, disk I/O performance can be substantially improved by employing parallel I/O schemes. Based on a distributed I/O architecture for parallel computers, we propose to use disk caches on several architectural levels, and confirm this by simulations of various structural options. In this paper, we describe the cache modelling approach and the I/O load model which has been derived from transaction-processing and general-purpose applications. Then we discuss the results for caches on single and multiple architecture levels. Large caches on I/O processors in combination with small caches on processing elements turn out to be the preferable structure. In addition, hardware caches can be employed at disk level for further performance improvement. For write operations, a delayed write strategy is shown to be superior to other modes.

Zusammenfassung

Während sich die Rechenleistung heutiger Computersysteme beständig erhöht, entwickelt sich die Platten-Ein/Ausgabe zu einem Engpaß. Dies wird besonders in hochparallelen Maschinen deutlich, wo das Mißverhältnis zwischen Rechen- und Ein/Ausgabeleistung häufig besonders groß ist. Diese Situation kann—in Analogie zur Multiprozessortechnologie—durch Einführung paralleler Ein/Ausgabe-Systeme entschärft werden. Ausgehend von einer solchen verteilten Ein/Ausgabe-Architektur für Parallelrechner, schlagen wir die Benutzung von Plattencaches auf verschiedenen Architekturebenen vor. Diese Cachestrukturen haben wir durch Simulationen näher untersucht. Wir beschreiben zunächst den Ansatz zur Cachemodellierung und das Lastmodell der Ein/Ausgabe, das von Transaktionsverarbeitung und General-Purpose-Lasten abgeleitet ist. Dann diskutieren wir die Ergebnisse für den Fall von Caches, die sich auf jeweils einer einzigen Architekturebene befinden, und für solche auf mehreren Architekturebenen gleichzeitig. Es ergibt sich, daß große Caches auf den Ein/Ausgabeprozessoren in Verbindung mit kleineren auf den Verarbeitungselementen die optimale Struktur darstellen. Zusätzlich können Hardwarecaches auf Plattenebene zur weiteren Leistungssteigerung eingesetzt werden. Bezüglich der Schreiboperationen vom Cache auf die Platte ergibt sich ein verzögerter Schreibmodus (delayed write) als beste Strategie.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Similar content being viewed by others

References

  1. Anon, et al.: A measure of transaction processing power. Datamation85, 112–118 (1985).

  2. Bhide, A., Stonebraker, M.: A performance comparison of two architectures for fast transaction processing. Proceedings Fourth International Conference on Data Engineering, Los Angeles, 1–5 Feb. 1988, pp. 536–545.

  3. Bohn, V., Härder, T., Rahm, E.: Extended memory support for high performance transaction systems. ZRI Report 5/90, University of Kaiserslautern, West Germany, 1990.

    Google Scholar 

  4. Eckardt, H.: Simulation of distributed sisk I/O for parallel computers. Siemens Internal Report No. SYS 3-BeG 034/91, 1991.

  5. Eckardt, H.: Performance evaluation of I/O architectures for transaction processing on parallel computers. PARS/PASA Workshop Proceedings of the Gesellschaft für Informatik, Paderborn, Germany 1992, pp. 73–77.

  6. Eckardt, H., Hellwagner, H., Klein, A.: Design of a scalable file system for distributed shared memory. PARS Workshop Proceedings of the Gesellschaft für Informatik, Dresden, Germany 1993, pp. 24–31.

  7. French, J. C., Pratt, T. W., Das, M.: Performance measurement of the Concurrent File System of the Intel iPSC/2 Hypercube, J. Par. Distr. Comp.17, 115–121 (1993).

    Article  Google Scholar 

  8. Ghosh, J., Goveas, K. D., Draper, J. T.: Performance evaluation of a parallel I/O subsystem for hypercube multicomputers, J. Par. Distr. Comp.17, 90–106 (1993).

    Article  Google Scholar 

  9. Härder, T., Rahm, E.: Multiprocessor database systems for high performance transaction systems (in German). Informationstechnik28, 214–225 (1986).

    Google Scholar 

  10. Hellwagner, H.: Design considerations for scalable parallel file systems, Comp. J.36, 741–755 (1993).

    Article  Google Scholar 

  11. Inmos, Ltd.: The T9000 transputer products overview manual. Inmos Ltd., document no. 72 TRN 228 00, 1991.

  12. Jenq, B. P., Twichell, B., Keller, T.: Locking performance in a shared nothing parallel database machine. Proceedings of the Fifth International Conference on Data Engineering, Los Angeles, 6–10 Feb. 1989, pp. 149–158.

  13. Lazowska, E. D., Zahorjan, J., Cheriton, D. R., Zwaenepoel, W.: File access performance of diskless workstations. ACM Trans. Comp. Syst.4, 238–268 (1986).

    Article  Google Scholar 

  14. Lieberman, D.: Parallel machine extends scalability to encompass I/O processing. Comp. Design, Nov. 1988, pp. 23–25.

  15. Makaroff, D. J., Eager, D. L.: Disk cache performance for distributed systems. Proceedings of the 10th International Conference on Distributed Computing Systems, Paris, May 28–June 1, 1990, pp. 212–219.

  16. Nelson, M. N., Welch, B. B., Ousterhout, J. K.: Caching in the Sprite network file system. ACM Trans. Comp. Syst.6, 134–154 (1988).

    Article  Google Scholar 

  17. Nitzberg, B., Lo, V.: Distributed shared memory: a survey of issues and algorithms. IEEE Comput.24, 52–60 (1991).

    Google Scholar 

  18. Ousterhout, J. K., Da Costa, H., Harrison, D., Kunze, J. A., Kupfer, M., Thompson, J. G.: A trace-driven analysis of the UNIX 4.2 BSD file system. Proceedings of the Tenth ACM Symposium on Operating System Principles, December 1985, pp. 15–24.

  19. Reddy, A. L. N., Banerjee, P.: Performance evaluation of multiple-disk I/O systems. 1989 International Conference on Parallel Processing, pp. I-315-I-318.

  20. Reddy, A. L. N., Banerjee, P.: An evaluation of multiple-disk I/O systems. IEEE Trans. Comp.38, 1680–1690 (1989).

    Article  Google Scholar 

  21. Reddy, A. L. N., Banerjee, P.: Design, analysis and simulation of I/O architectures for hypercube multiprocessors. IEEE Trans. Par. Distr. Syst.1, 140–151 (1990).

    Article  Google Scholar 

  22. Reyzl, E., Eckardt, H.: Performance evaluation for high-performance interconnection networks. Proceedings of the 11th ITG/GI Meeting, Munich, March 1990, pp. 275–287.

  23. Smith, A. J.: Disk cache-miss ratio analysis and design considerations. ACM Trans. Comp. Syst.3, 161–203 (1985).

    Article  Google Scholar 

Download references

Author information

Authors and Affiliations

  1. Corporate Research and Development, ZFE ST SN 1, Otto-Hahn-Ring 6, D-81739, München, Federal Republic of Germany

    H. Eckardt

Authors
  1. H. Eckardt
    View author publications

    You can also search for this author in PubMed Google Scholar

Rights and permissions

Reprints and permissions

About this article

Cite this article

Eckardt, H. Disk cache architectures for transaction-like applications in parallel computers. Computing 53, 13–31 (1994). https://doi.org/10.1007/BF02262106

Download citation

  • Received:

  • Revised:

  • Issue Date:

  • DOI: https://doi.org/10.1007/BF02262106

Key words

Search

Navigation