When Piecewise Determinism Is Almost TrueEdith

Most existing log-based recovery techniques assume perfect piecewise determinism. In practice, however, the behavior of certain events is determined by the execution environment, and is not replayble. In this paper , we introduce a probabilistic piecewise determinis-tic model, and develop a probability tracking mechanism to achieve high-conndence output commits. The paper also addresses the issue of scalability by proposing an eecient on-line algorithm that accurately estimates the probability of determinism by piggybacking only constant-size vectors.

[1]  David B. Johnson,et al.  Efficient transparent optimistic rollback recovery for distributed application programs , 1993, Proceedings of 1993 IEEE 12th Symposium on Reliable Distributed Systems.

[2]  Leslie Lamport,et al.  Distributed snapshots: determining global states of distributed systems , 1985, TOCS.

[3]  Yennun Huang,et al.  Software Implemented Fault Tolerance Technologies and Experience , 1993, FTCS.

[4]  A. Prasad Sistla,et al.  Efficient distributed recovery using message logging , 1989, PODC '89.

[5]  Willy Zwaenepoel,et al.  Manetho: Transparent Rollback-Recovery with Low Overhead, Limited Rollback, and Fast Output Commit , 1992, IEEE Trans. Computers.

[6]  James R. Russell,et al.  Optimistic failure recovery for very large networks , 1991, [1991] Proceedings Tenth Symposium on Reliable Distributed Systems.

[7]  Andreas Reuter,et al.  Transaction Processing: Concepts and Techniques , 1992 .

[8]  Edith Cohen,et al.  Estimating the size of the transitive closure in linear time , 1994, Proceedings 35th Annual Symposium on Foundations of Computer Science.