L. Huston and P. Honeyman. Peephole log optimization. In Proc. Workshop on Mobile Computing Systems and Applications, Santa Cruz, CA, Dec. 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....disconnected) As well, delays in the reintegration of file changes with the home file server (write backs) can lead to consistency problems. Support for disconnected and weakly connected operation has been studied by researchers at Carnegie Mellon University [3] and the University of Michigan [4], who have shown that even a weak connection can permit enough processing of read misses and write backs to maintain operability while mobile. Adapting file system support so that it can be made available to mobile clients across the full range of connection possibilities is the overall goal of ....

....In this way, the final state of the file system is the same as if the changes were made while connected, although considerable time may have elapsed since the changes were actually made. Log files can be optimised to save disk space at the client and to decrease the time needed for reintegration [4, 7]. This optimisation is accomplished through recognition of which operations actually affect the final state of the file system. For example, if the same part of a file is updated more than once, only the last update needs to be kept in the log, since the previous changes are overwritten by ....

L. Huston and P. Honeyman. Peephole log optimization. In Proc. Workshop on Mobile Computing Systems and Applications, Santa Cruz, CA, Dec. 1994.

....[42] Two optimizations before a new record is appended: 1) any operation which overwrites the effect of earlier operations may cancel the corresponding records (2) an inverse operation (e.g. rmdir) cancels both the inverting and inverted (e.g. mkdir) log records. The Little Work project [31] suggests applying rule based techniques used in compiler peephole optimizers. Use such an off the shelf optimizer. Optimization at a preprocessing step before reintegrating at reconnection. Two types of rules: 1) replacement rules remove adjacent redundant operations, e.g. a create followed by ....

L. Huston and P. Honeyman. Peephole Log Optimization. In Proceedings of the IEEE Workshop on Mobile Computing Systems and Applications, Santa Cruz, CA, December 1994.

....dynamic sets, applications had to be changed and they have to be able to tolerate the reordering of requests implicit in the use of dynamic sets. Further information on the Little Work Project, an approach in contrast to the Coda System, which applies optimizations incrementally, is presented in [58]. The primary advantage is that the optimization code is a separable component; hence it is easy to apply to multiple file systems. Predictive caching techniques are elaborated in [68] The goal of this work is to reduce the burden on users of specifying files to be hoarded in anticipation of ....

....for example from an ethernet connection to a wireless modem, without rebooting or restarting applications. In [74] the authors discusses a client side only approach to providing access to a a heterogenous set of network file services. This is different from the approach in systems such as Coda [58] which rely on changes to both the client and the server. Rather than committing to a specific file access protocol, they support access to multiple file services. In [77] the authors present a design for supporting file access in a mobile environment. The main design goals they address are to ....

L. Huston and P. Honeyman, Peephole Log Optimization, Proceedings of the IEEE Workshop on Mobile Computing Systems and Applications, 1994.

.... the limit, if reconciliation never occurs, no conflicts will be detected regardless of the degree of concurrent updates (This effect is exacerbated by the the common optimization of compressing multiple changes to the same object into a single update to save time and space during reconciliation [3, 5, 6]. A second drawback with conflict counts is the difficulty of measuring conflicts in systems with more than two replicas [1] The measurement can depend on the pattern of both updates and reconciliations. Consider four replicas of a single object. Replicas 1 and 3 receive independent updates ....

L. B. Huston and P. Honeyman. Peephole log optimization. In Proceedings of the Workshop on Mobile Computing Systems and Applications 1994, Santa Cruz, CA, Dec. 1994.

.... never occurs, no conflicts will be detected regardless of the degree of concurrent updates (This effect is exacerbated by the optimizations done in most practical systems, in which multiple changes to the same object are compressed into a single update to save time and space during reconciliation [3, 5, 6]. A second drawback with conflict counts is the difficulty of measuring conflicts in systems with more than two replicas [1] We have found that the number of measured conflicts can depend on the pattern of both updates and reconciliations. As a simple example, consider four replicas of a single ....

L. B. Huston and P. Honeyman. Peephole log optimization. In Proceedings of the Workshop on Mobile Computing Systems and Applications 1994, Santa Cruz, CA, Dec. 1994.

....to the file server all file system changes that were made. In this way, the final state of the file system is (ideally) the same as if the changes were made while connected. Log files are usually optimised in order to save disk space at the client and to decrease the time needed for reintegration [18, 46]. This optimisation is accomplished through recognition of which operations actually affect the final state of the file system. Operations that do not change the state of the file system (such as reads) are called non mutating, and are not recorded in the log file. Mutating operations are ....

L. Huston and P. Honeyman. Peephole log optimization. In Proc. Workshop on Mobile Computing Systems and Applications, Santa Cruz, CA, Dec. 1994.

....in the background. The first is rational management of network resources, to prevent response times for interactive and other traffic from suffering. The second issue is the effect on optimization: we and our CODA counterparts have observed that optimization of large logs can be considerable [7, 18], vastly reducing the amount of network traffic necessary for replay. Aggressive background replay may deny us this savings. 5.1. Priority queuing The network is a primary resource in the mobile environment, so it is vital to keep replay traffic from interfering with a user s other work. Studies ....

....to propagate updates aggressively. On the other hand, delaying replay offers an opportunity for optimizing the log. Ousterhout reports that most UNIX files have a lifetime under three minutes and that 30 40 of modified file data is overwritten within three minutes [17] Using our optimizer [7], we find it typical for 70 of the operations in a large log to be eliminated. In fact, the larger the log, the greater the fraction of operations eliminated by the 4 Partially Connected Operation optimizer. It is clear that delaying log replay can help reduce the amount of data propagated ....

L.B. Huston and P. Honeyman, "Peephole Log Optimization," pp. 25-32 in Bulletin of the Tech. Comm. on Op. Sys. and Appl. Env., IEEE Computer Society (Spring 1995).

....in the background. The first issue is rational management of network resources, so that the response times for interactive and other traffic do not suffer. The second issue is the effect on optimization: we and our CODA counterparts have observed that optimization of large logs can be considerable [8, 17], vastly reducing the amount of network traffic necessary for replay. Aggressive background replay may deny us this savings. 5.1. Priority queuing The network is a primary resource in the mobile environment, so it is vital to keep replay traffic from interfering with a user s other work. ....

....logged operations aggressively. On the other hand, delaying replay offers an opportunity for optimizing the log. Ousterhout reported that most UNIX files have a lifetime under three minutes and that 30 40 of modified file data is overwritten within three minutes [16] Using our optimizer [8], we find it typical for 70 of the operations in a large log to be eliminated. It is clear that delaying log replay can significantly reduce the amount of data propagated to the server. We may wish to enforce a minimum delay before replaying an operation, especially on networks with a per packet ....

L.B. Huston and P. Honeyman, "Peephole Log Optimization," in Proc. of the IEEE Workshop on Mobile Computing Systems and Applications, Santa Cruz (December 1994).

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