Parallelizing I/O intensive applications for a workstation cluster: a case study

ABSTRACT

An I/O intensive application, parallel full text retrieval based on a signature file method, is studied. The text retrieval system is implemented on a cluster of DEC5000 workstations connected by Ethernet. Experiments are performed to evaluate the benefit and cost for running such an application in a workstation cluster. Results show that substantial improvement in speed can be obtained through parallelism in disk accesses, despite the high communication and synchronization overhead that is incurred. Several factors that affect the performance of a parallel I/O application in this type of computing environment are discussed. The advantages of a workstation cluster are its large combined I/O buffer capacity, and possible concurrent accesses to disks local to each workstation. Our study demonstrates that these advantages, when exploited properly, can lead to effective performance improvement without the need for additional hardware.

REFERENCES

• 1: S. Christodoulakis , F. Ho , M. Theodoridou, The multimedia object presentation manager of MINOS: a symmetric approach, Proceedings of the 1986 ACM SIGMOD international conference on Management of data, p.295-310, May 28-30, 1986, Washington, D.C., United States
• 2: [2] A.L. Cheung and A.P. Reeves, "High Performance Computing on a Cluster of Workstations", Proceedings of the First International Symposium on High-performance Distributed Computing, Syracuse, New York, Sept, 1992.
• 3: C. Faloutsos, Signature-based text retrieval methods: a survey, Data Engineering, v.13 n.1, p.25-32, Mar. 1990
• 4: [4] G.A. Geist, "Network Based Concurrent Computing on the PVM System", Technical report TM-11826, Oak Ridge National Lab., 1991.
• 5: Z. Lin , C. Faloutsos, Frame-Sliced Signature Files, IEEE Transactions on Knowledge and Data Engineering, v.4 n.3, p.281-289, June 1992 [doi>10.1109/69.142018]
• 6: Zheng Lin, Concurrent frame signature files, Distributed and Parallel Databases, v.1 n.3, p.231-249, July 1993 [doi>10.1007/BF01263332]
• 7: [7] M. Psrashar and S. Hariri, "A Requirement Analysis for High Performance Distributed Computing over LAN's", Proceedings of the First International Symposium on High-performance Distributed Computing, Syracuse, New York, Sept, 1992.
• 8: [8] J. Price, "The Optical Disk Pilot Project at the Library of Congress", Video-disc and Optical Disk, vol. 4, no. 6, pp. 424-432, Nov. 1984.
• 9: R. Sacks-Davis , A. Kent , K. Ramamohanarao, Multikey access methods based on superimposed coding techniques, ACM Transactions on Database Systems (TODS), v.12 n.4, p.655-696, Dec. 1987 [doi>10.1145/32204.32222]
• 10: C. J. Van Rijsbergen, Information Retrieval, Butterworth-Heinemann, Newton, MA, 1979

INDEX TERMS

Primary Classification:
  F. Theory of Computation
  F.1 COMPUTATION BY ABSTRACT DEVICES
      F.1.2 Modes of Computation
          Subjects: Parallelism and concurrency

Additional Classification:
  C. Computer Systems Organization
  C.5 COMPUTER SYSTEM IMPLEMENTATION
      C.5.3 Microcomputers
          Subjects: Workstations

  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL

General Terms:
Design, Experimentation, Management, Measurement, Performance, Theory