H. Let and D. Duchamp. Transparent File Prefetching. In preparation, Columbia University, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....in the file access sequence. It is not yet clear how effective this method will be. One known restriction is the inability to separate file accesses per formed by concurrently executing processes. 2 Our Approach Our approach is based on two of the authors previ ous work: in file prefetching [13, 12, 8]. Both hoarding and preftching involve anticipatory file fetches: bringing files from a remote server into a local cache before they are needed. The goals of these two techniques are quite different, however. Prefetching is an online technique that is mainly concerned with improving performance: ....

H. Let and D. Duchamp. Transparent File Prefetching. In preparation, Columbia University, 1995.

....could predict the future exactly, it must be rewritten to obtain the benefits of prefetching. Finally, allowing applications to control or influence the file cache means the system must provide mechanisms and policies to ensure fairness among applications [CFKL95] Duchamp et al. TD91, LD95] use an automatic approach based on stacks of access trees that correspond to process trees and the files referenced by the process tree. The operating system monitors the current access tree, constantly looking for matches against past trees. As soon as the operating system identifies a match, ....

Hui Lei and Dan Duchamp. Transparent file prefetching. Technical report, Dept. of Computer Science, Columbia University, March 1995. submitted for publication.

....of predicting future file system requests, has prevented prefetching from producing any significant performance improvements in LAN based distributed file systems. To date, few mobile file systems have implemented any prefetching mechanism, and research in this area is sparse. Lei and Duchamp [13] have proposed a file based prefetching policy for use with mobile clients. The approach taken is to store file reference patterns in tree structures, and then attempt to match current references with a stored pattern. As soon as a match is made, the remaining files in the pattern are prefetched. ....

Lei, H., and Duchamp, D. Transparent file prefetching. Technical report, Department of Computer Science, Columbia University, New York, NY, 1995.

....is similar to the file working sets approach in that it leverages information about files that have already been accessed. The Coda File System works well when users cooperate by updating the hoard database (HDB) However, relying on information provided by users can lead to two types of problems [23]. First, many users may refuse to do the necessary work to keep the HDB primed with the correct contents. Second, users may be unable to describe fully what should be in the HDB. For example, users may be unaware of files required by the implementation of an application Both ideas presented by ....

H. Lei and D. Duchamp, "Transparent file prefetching." Submitted for publication, Mar. 1995.

....not because there is insufficient bandwidth to meet demand, but because there is insufficient bandwidth at any instant. Prefetching algorithms are significantly more complex than other cache management algorithms since they attempt to predict future requests before they occur. Previous work [10, 23, 28] has relied on keeping extended histories of a user s file access patterns, and then recognising these patterns when they occur again. While some success has been achieved with these algorithms, it may also be possible to extend an algorithm so that it is able to successfully predict file accesses ....

....that is, the server notifies the client when a modified file should be written back (based on when it is convenient for the server) The server also informs the client when a modified file may be removed from its cache. Work has also been done on a prefetching mechanism for the client cache [28]. The approach taken is to store file reference patterns in tree structures, and then attempt to match current references with a stored pattern. As soon as a match is made, the remaining files in the pattern are prefetched. Once files have been prefetched, they are treated the same as ....

H Lei and D. Duchamp. Transparent file prefetching. Technical report, Department of Computer Science, Columbia University, New York, NY, 1995.

....in the file access sequence. It is not yet clear how effective this method will be. One known restriction is the inability to separate file accesses performed by concurrently executing processes. 2 Our Approach Our approach is based on two of the authors previous work in file prefetching [13, 12, 8]. Both hoarding and prefetching involve anticipatory file fetches: bringing files from a remote server into a local cache before they are needed. The goals of these two techniques are quite different, however. Prefetching is an online technique that is mainly concerned with improving performance: ....

H. Lei and D. Duchamp. Transparent File Prefetching. In preparation, Columbia University, 1995.

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