T. Park, S.B. Cho, and H.Y. Yeom. An Efficient Logging Scheme for Recoverable Distributed Shared Memory Systems. In Proc. of the 17th Int'l Conf. Distributed Computing Systems, pp. 305--313, May 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....are discussed in [24] for a uniprocessor and in [23] for a cluster of machines. In coordinated checkpointing [28, 15] the system checkpoints the state of all threads simultaneously by stopping computation periodically. In order to avoid expensive checkpointing operations, some designs [21, 22, 8] checkpoint infrequently while logging changes to shared memory between successive checkpoints. In [8] Costa et al. describe a logging and checkpointing scheme for lazy release consistency. We extend this work by identifying properties that any highly available software DSM using release ....

T. Park, S. B. Chou, and H. Y. Yeom. An efficient logging scheme for recoverable distributed shared memory. In Proc. of the 17th Int'l Conf. on Distributed Computing Systems (ICDCS-17), May 1997.

....has accessed into the stable storage in the access order, so that the process can retrieve the logged messages at the same computational points after the failure recovery. However, these schemes could incur high logging overheads during the failure free operations. The writer based logging schemes [6, 7] utilize the volatile log of the writer process for each data item, and the corresponding readers only log the access information into the stable storage. As a result, the logging overhead can be reduced, however, the overhead of stable logging is still non negligible. One solution suggested in ....

....by accessing the common data page, and Definition 1. d) states that the dependency relation is transitive. 4 Such dependency relation between the state intervals may cause possible inconsistency problems during the rollback recovery. Figure 1 shows a typical example of inconsistent recovery case [6]. The notations, R(X) and W (X) in the figure represent the read and the write operations on a data page X, respectively, and the notations U(A) and L(A) represent the release and the acquire operations on a lock A, respectively. Now, suppose that process p i should roll back to its latest ....

T. Park, S.B. Cho, and H.Y. Yeom. An Efficient Logging Scheme for Recoverable Distributed Shared Memory Systems. In Proc. of the 17th Int'l Conf. Distributed Computing Systems, pp. 305--313, May 1997.

....same computation by replaying the logged data items; that is, the rollback of one process does not affect other processes. However, since the logging itself may cause nonnegligible overhead, many solutions to reduce the logging overhead have been suggested for the sequentially consistent DSM system[4, 7, 11]. The memory consistency model of the DSM system is an important factor to characterize the inter process dependency. Hence, for the lazy release consistency(LRC) model, which is one of the relaxed memory models[1, 3, 5] other requirements are needed to reduce the logging overhead. In [11] the ....

....b) 2 Pi Pj Failure X R(X1) Ci Cj W(X1) U(A) X1 L(A) Figure 1. An Inconsistent Recovery Line Such dependency relation between the state intervals may cause possible inconsistency problems during the rollback and recovery. Figure 1 shows a typical example of inconsistent rollback recovery case[7]. The notations, R(X 1 ) and W (X 1 ) in the figure represent the read and the write operations on a data page X 1 , respectively, and the notations U(A) and L(A) represent the release and the acquire operations on a lock A, respectively. Now, suppose the process p i in Figure 1 rolls back to ....

[Article contains additional citation context not shown here]

T. Park, S.B. Cho, and H.Y. Yeom. An efficient logging scheme for recoverable distributed shared memory systems. In Proc. of the 17th Int'l Conf. Distributed Computing Systems, May 1997.

....system [27] To further reduce the logging overhead, in [19] the logging of the accessed data value and its usage duration is delayed until the value is invalidated in the system, and in [11] the data value is logged when it is written instead of being logged each time when it is accessed. In [20], the logging overhead is substantially reduced by logging the data value at the site where it is produced and at the time when the value is invalidated in the system. For the lazy release consistent DSM system, it is suggested in [27] that the process should make a stable log of the accessed ....

T. Park, S.B. Cho, and H.Y. Yeom. An efficient logging scheme for recoverable distributed shared memory systems. In Proc. of the 17th Int'l Conf. Distributed Computing Systems, 1997.

....approach[10] Both approaches ensure no domino effect since each checkpoint establishes a consistent recovery line, however, the overhead caused by checkpointing coordination may severely degrade the system performance. Another approach to avoid the cascading rollback is to use message logging [3, 7, 9, 5]. Each process in this approach logs the accessed data values on a stable storage in addition to the checkpoints taken independently. Hence, the process can reproduce the same sequence of computation after a rollback by recomputing with the logged data values. As a result, the process can locally ....

T. Park, S. Cho, and H. Yeom, "An efficient logging scheme for recoverable distributed shared memory systems", In Proc. of the 17th Int'l Conf. on Distributed Computing Systems, 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC