Data base recovery at CMIC

ABSTRACT

Several techniques have been combined to provide for data base recovery at CMIC. The CMIC environment is presented first in order to identify the constraints which data base recovery must satisfy.A technique is described for updating mass-storage structures (a B-tree in this case) in such a way that all information already addressable through the mass-storage structure can always be addressed through the mass-storage structure, even while the mass-storage structure is being updated. Audit trail recovery points are defined to be records on the audit trail tape with the property that all preceeding data base updates have been written to mass storage. Because of the mass-storage updating technique, any data base update not addressable through the data base mass-storage B-tree could not have been written to mass storage. Therefore, by the definition of a recovery point, the update must have occurred after the most recent recovery point.The audit trail is always updated before the data base. In the event of a failure which does not destroy mass storage (a soft crash), the data base mass-storage allocation is recovered by a tree walk through the B-tree ("hot start"). Then any updates which occurred after the last recovery point on the audit trail are restored from the audit trail.Finally, the problem of program restart after data base recovery is discussed. The question of whether deadlock is a reasonable price to pay for data base rollback is explained, but not answered. This question will have to be addressed as more data base systems try to support concurrent transaction and batch updates, distributed data bases, or data bases shared by independent host computers.

REFERENCES

• 1: {Bayer 72} R. Bayer and E. M. McCreight "Organization and Maintenance of Large Ordered Indexes", Acta Informatica 1, 3 (1972) 173-189.
• 2: {Bjork 75} L. A. Bjork, Jr., "Generalized Audit Trail Requirements and Concepts for Data Base Applications", IBM Systems Journal 14, 3 (1975), 229-245.
• 3: {DCU 75} "Statewide CRT Network Cuts Paperwork in Managing Personnel", The Data Communications User, May, 1975, 79-80.
• 4: {Giordano 75} Nicholas J. Giordano and Laurence B. LeGore, Systems Specification for a Backward Recovery Procedure for the FMS-8 Data Base, CMIC Internal Report, April 8, 1975.
• 5: {Keehn 74} D. G. Keehn and T. O. Lacy, "VSAM Data Set Design Parameters", IBM Systems Journal 13, 3 (1974), 186-212.
• 6: Donald E. Knuth, The art of computer programming, volume 3: (2nd ed.) sorting and searching, Addison Wesley Longman Publishing Co., Inc., Redwood City, CA, 1998
• 7: B. Randell, System structure for software fault tolerance, Proceedings of the international conference on Reliable software, p.437-449, April 21-23, 1975, Los Angeles, California
• 8: Marvin Stuart Schwartz, A storage hierarchial addressing-space for a computer file system., 1973
• 9: {Univac 72} FMS-8 Design Specifications, Sperry Univac, May 1972.
• 10: {Univac 74A} Sperry Univac 1100 Series Data Management System (DMS-1100) Schema Definition, Sperry Univac publication UP-7907 Rev. 2, 1974, Sections 3.2.8 and 3.3.8.
• 11: Univac 74B Sperry Univac 1100 Series Data Management System (DMS-1100), American National Standard COBOL (ASCII) Data Manipulation Language, Sperry Univac publication UP 7992 Rev. 1, 1974, Section 3.3.11, and Chapter 4.
• 12: Univac 75 Sperry Univac 1100 Series Data Management System (DMS-1100) System Support Functions, Sperry Univac publication UP7909 Rev. 3, 1975, Chapter 6.
• 13: USE 72 "Panic Dump Analyzer", USE Conference Notes, March, 1972.

INDEX TERMS

Keywords:
"hot start", B-tree, Leaf-First Rule, VSAM, audit trail, data base recovery, hard crash, on-line data base, program restart, reload and restore, soft crash, track splitting, transaction