Kim Korner. Intelligent caching for remote file service. In Proceedings of the Tenth International Conference on Distributed Computer Systems, pages 220--226. IEEE Society Press, 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....will not) be accessed in the future. It also mentioned a static analysis approach to adding such hints to looping array codes automatically. Another proposal is to use the name and directory of a file to predict how the file will be accessed and select a cache ejection policy for the file s data [25]. One advantage of improving the cache ejection policy relative to I O prefetching is that it may also reduce the amount of data that must be fetched from disk. One disadvantage is that it cannot hide the I O latency of fetching data that is not cached. Another way to reduce I O stall time, which ....

Kim Korner. Intelligent caching for remote file service. In Proceedings of the 10th International Conference on Distributed Computing Systems (ICDCS), 1990.

....replacement policies, is another such remedy. Choosing the correct cache replacement policy in an array can improve its performance [19, 21, 30, 35, 48] Some studies suggest using least frequently used [15, 46] or frequency based [30] replacement policies instead of LRU in file servers. MRU [23] or next block prediction [29] policies have been shown to provide better performance for sequential loads. LRU or clocking policies [10] can yield acceptable results for database loads; for example, the IBM DB2 database system [36] implements an augmented LRU style policy. Our DEMOTE operation ....

K. Korner. Intelligent caching for remote file service. In Proc. of the 10th Intl. Conf. on Distributed Computing Systems, pages 220--226. May 1990.

....replacement decisions. If they cannot perform better than RAND, then there is almost certainly some underlying flaw in the decision strategy. In effect, RAND, provides a kind of lower bound on performance. That is, there is no reason to use any policy which performs worse than RAND. Korner [3] investigates an alternative cache management policy for distributed fileservers. His algorithm chooses among different replacement strategies (LRU, MRU, or no caching) for different classes of blocks depending upon the anticipated reference patterns. Performance improvements of up to 340 were ....

....representative of the typical workload in our installation they are fairly short. Future work would benefit from larger traces (measured in days rather than hours) and from traces gathered in industrial environments. Third, more sophisticated fileserver cache management strategies such as Korner s [3], but with an underlying frequency based approach rather than an LRU approach would be worthwhile investigating. 6 Acknowledgements The authors would like to thank Michael Molloy, from Hewlett Packard, for providing technical assistance and several tools used to instrument the HP UX 8.0 ....

Korner, K., "Intelligent Caching for Remote File Service," ICDCS, May 1990, pp. 220-226.

.... is used to guide non sequential prefetching within a file [42] Fido is an example of a predictive cache that prefetches by using an associative memory to recognize access patterns over time [62] Knowledge based caching has been proposed to enhance cache performance of remote file servers [37]. Mowry et al. [56] propose a method for using input output access patterns determined by the compiler to hide input output latency by prefetching, assuming the application relies upon paged virtual memory for all input output. These studies focus on application input output, but the method can ....

Korner, K. Intelligent Caching for Remote File Service. In Proceedings of the 10th International Conference on Distributed Computing Systems (May 1990), pp. 220--226.

....of accessing slow mass storage, even small improvements in hit rate are important. We plan to develop and evaluate caching algorithms tailored to the mass storage and low speed network applications as well as studying caching algorithms that have been suggested by others. Willick [21] and Korner [22] investigate alternatives to LRU caching for a distributed file system environment. We may perform simulations using their suggested caching schemes in the context of multilevel caching. However, it is difficult to guess the effectiveness of these caching policies for our Formatted October 14, ....

K. Korner, "Intelligent Caching for Remote File Service," pp. 220-226 in Proc. 10th Intl. Conference on Distributed Computing Systems, Paris, France (May, 1990).

....when it has been closed on all clients. Experiments with file operations indicate that under certain conditions, client caches allow diskless Sprite workstations to perform almost as well as clients with disks. In addition, client caching reduces server load by 50 and network traffic by 75 . In [48], Korner suggested the use of intelligent methods to improve the effectiveness of caching. Caching algorithms using higher level knowledge can generate expectations of user process behavior to provide hints to the file system. Using Unix based generalizations of file usage by programs, depending ....

K. Korner. Intelligent Caching for Remote File Service. In Proceedings of the 10th IEEE International Conference on Distributed Computing Systems, Paris, France, June 1990.

....long sequential runs. Others, notably Kotz, have looked at detecting more complex access patterns and prefetching non sequentially within a file [Kotz91] At the level of whole files or database objects, a number of researchers have looked at inferring future accesses based on past accesses [Korner90, Kotz91, Tait91, Palmer91, Curewitz93, Griffioen94]. The danger in speculative prefetching based on historical access patterns is that it risks hurting, rather than helping, performance [Smith85] As a result of this danger, speculative prefetching is usually conservative, waiting until its theories are confirmed by some number of demand accesses. ....

Korner, K., "Intelligent Caching for Remote File Service, Proc. of the Tenth Int. Conf. on Distributed Computing Systems, 1990, pp.220-226.

....relationships between files has also been a significant research topic. Fido is an example of a predictive cache that prefetches by using an associative memory to recognize access patterns over time [21] Knowledge based caching has been proposed to enhance cache performance of remote file servers [11]. Some approaches use probabilistic methods to create models of user behavior to guide prefetching [15, 4, 13] This work is similar in spirit to our HMM classification methodology. The difficulty with modeling user behavior probabilistically is that recently accessed files are more likely to be ....

Korner, K. Intelligent Caching for Remote File Service. In Proceedings of the 10th International Conference on Distributed Computing Systems (May 1990), pp. 220--226.

....of accessing slow mass storage, even small improvements in hit rate are important. We plan to develop and evaluate caching algorithms tailored to the mass storage and low speed network applications, and to study caching algorithms that have been suggested by others. Willick et al. 27] and Korner [28] investigate alternatives to LRU caching for a distributed file system. We may perform simulations using their suggested caching schemes in the context of multilevel caching. However, it is difficult to guess the effectiveness of these caching policies for our use, because both papers use short ....

K. Korner, "Intelligent Caching for Remote File Service," pp. 220-226 in Proc. 10th Intl. Conference on Distributed Computing Systems, Paris (May, 1990).

.... Smi85] In the context of mobile computing, long term (batch) file prefetching has been used to support disconnected operation [Sat90, SK93, Kue94, KPR94, ABC89, HH93] File system prefetching has long been used to improve intrafile accesses (accesses within a file) Smith [Smi78] and Korner [Kor90] examined the problem of predicting future file block accesses within the current file. Such models can significantly improve access to large files containing many blocks. Kotz et al. KE90] examined the problem of intrafile prefetching in the context of scientific applications and parallel file ....

K. Korner. Intelligent Caching for Remote File Service. In the Proceedings of the 10th Inernational Conference on Distributed Computing Systems, pages 220--226, May 1990.

.... and Galley from Nieuwejaar [22] and implicit policy control, via hints [4] expressive user interfaces (e.g. ELFS [7] and collective input output as in del Rosario [23] and Kotz [24] or intelligent modeling of file access (e.g. Fido from Palmer [25] and knowlege based caching from Korner [26]) Fido is an example of a predictive cache that prefetches by using an associative memory to recognize access patterns over time. Knowledge based caching has been proposed to enhance cache performance of remote file servers. The second component of our research, dynamic performance based ....

KORNER, K. Intelligent Caching for Remote File Service. In Proceedings of the 10th International Conference on Distributed Computing Systems (May 1990), pp. 220-226.

.... that are used to guide non sequential prefetching within a file [12] Fido is an example of a predictive cache that prefetches by using an associative memory to recognize access patterns over time [18] Knowledge based caching has been proposed to enhance cache performance of remote file servers [9]. Our work uses trained neural networks and statistical models to classify global access patterns, and can be trained to recognize new patterns as the need arises. 7 Conclusions Global access pattern information is critical to optimizing input output performance of parallel applications. ....

Korner, K. Intelligent Caching for Remote File Service. In Proceedings of the 10th International Conference on Distributed Computing Systems (May 1990), pp. 220--226.

....both time and space requirements for the algorithm are modest. Ideally, one might wish to be able to prefetch only portions of a file. It is beyond the scope of our algorithm to handle this case; however, it is an interesting area for future research, with preliminary work already done by Korner [25]. 1.3 Outline of the Dissertation Chapter 2 describes previous research in distributed file systems, providing a background for our own work. Chapter 3 describes the algorithms involved in efficient variable consistency replication, including experimental data, comparison with existing systems, ....

....and the cache sizes in the simulations are 108 extremely small: only 10 pages, with a brief comment that a (still small) cache of 50 pages produces similar results. The work most closely related to our own is Korner s work on detecting and exploiting the block access patterns of individual files [25]. Several important aspects make Korner s work different. First, the detection of patterns is an off line process performed by two expert systems, whereas our method provides on line real time detection and exploitation. Further, whereas our method is an adjustment to LRU, many of the access ....

K. Korner. Intelligent Caching for Remote File Service. In Proc. Tenth Intl. Conf. on Distributed Computing Systems, pages 220--226. IEEE, May 1990.

....4 data in the latter portions of the block. More explicitly, the file system can readahead sequential blocks of a file [Smith85, Feiertag71] Recently, other researchers have explored prefetching for more complex access patterns by inferring future requests from a user s I O request stream [Kotz91, Tait91, Palmer91, Korner90]. For well behaved applications, these techniques can effectively prefetch data and reduce cache misses. However, for the many applications whose access patterns appear random or that touch data only once, such techniques are ineffective. The greater issue, though, is how well these techniques ....

....increase throughput that TIP uses to reduce read latency and application execution time. Finally, TIP uses its knowledge of future I O requests to improve cache management and reduce cache miss ratios. It is often possible to outperform an LRU page replacement algorithm, even without prefetching [Chou85, Korner90]. Unneeded blocks can be released early and needed blocks can be held longer. For example, knowing that a large file will be read sequentially twice, a TIP cache manager could hold onto the first few blocks of the file for use at the start of the second access while immediately flushing others and ....

[Article contains additional citation context not shown here]

Korner, K., "Intelligent Caching for Remote File Service, Proc. of the Tenth Int. Conf. on Distributed Computing Systems, 1990, pp.220-226.

....long sequential runs. Others, notably Kotz, have looked at detecting more complex access patterns and prefetching non sequentially within a file [Kotz91] At the level of whole files or database objects, a number of researchers have looked at inferring future accesses based on past accesses [Korner90, Kotz91, Tait91, Palmer91, Curewitz93, Griffioen94]. The danger in speculative prefetching based on historical access patterns is that it risks hurting, rather than helping, performance [Smith85] As a result of this danger, speculative prefetching is usually conservative, waiting until its theories are confirmed by some number of demand accesses. ....

Korner, K., "Intelligent Caching for Remote File Service, Proc. of the 10th Int. Conf. on Distributed Computing Systems, 1990, pp.220-226.

.... that are used to guide non sequential prefetching within a file [12] Fido is an example of a predictive cache that prefetches by using an associative memory to recognize access patterns over time [18] Knowledge based caching has been proposed to enhance cache performance of remote file servers [9]. Our work uses trained neural networks and statistical models to classify global access patterns, and can be trained to recognize new patterns as the need arises. 7 Conclusions Global access pattern information is critical to optimizing input output performance of parallel applications. ....

Korner, K. Intelligent Caching for Remote File Service. In Proceedings of the 10th International Conference on Distributed Computing Systems (May 1990), pp. 220--226.

....contact with a system file volume. The client cache is large enough to accommodate the whole volume of its user(s) as well as any other needed files. 16 The work most closely related to our own is Korner s work on detecting and exploiting the block access patterns of individual files [6]. The style of this work is similar to ours in that patterns are detected and exploited. However, several important aspects make Korner s work different: 1. The detection of patterns is an off line process performed by two expert systems, whereas our method provides on line real time detection ....

K. Korner. Intelligent Caching for Remote File Services. In Proc. Tenth Intl. Conf. on Distributed Computing Systems, pages 220-226. IEEE, May, 1990.

.... measured the effectiveness of the LRU replacement algorithm [Bel66] Since then, sequential readahead has become a widely used technique [FO71, MJLF84] Others have shown how to extract much more complex access patterns from an examination of past accesses [LD97, GA95, CKV93, KE93, PZ91, TD91, Kor90] Unfortunately, such speculative prefetching risks hurting, rather than helping, performance [Smi85] As a result, history based prefetching must be conservative and need not address the resource management issues related to very deep prefetching that are the focus of this paper. Sometimes an ....

Kim Korner. Intelligent caching for remote file service. In Proceedings of the 10th Intl. Conf. on Distributed Computing Systems, pages 220--226, 1990.

....of sequential reads in the Unix fast file system[16] and to increase availability in distributed file systems[9] to name just a few uses. Two forms of prefetching exist: inferential and informed. In the former, the system bases predictions on observations of past references. For instance, Korner[10] proposes off line analysis of file traces to establish rules for mapping file types to access patterns, and using the rules as hints to drive the preloading of data into the buffer cache. Kotz and Ellis[11] propose on line detection of sequential access to portions of a file to drive ....

K. Korner. Intelligent caching for remote file service. In Proceedingsof the 10th International Conference on Distributed Computing Systems, 1990.

....buffer cache design are different. Most notably, the accesses are inherently at a higher level (i.e. files and logical blocks) whereas the disk caches operate at a lower level (i.e. physical blocks) The issue of using application semantics to dictate replacement algorithms is explored in [27]. In their scheme, buffers are partitioned, and different CHAPTER 3. RELATED WORK 26 schemes are employed for each partition. Their work differs from ours in that they take a knowledge base approach to the problem, where they infer the optimal strategy not through system calls but through file ....

Korner, K. Intelligent caching for remote file service. In Proceedings of the IEEE International Conference on Distributed Computing Systems, pages 220--226, 1990.

.... that are used to guide non sequential prefetching within a file [11] Fido is an example of a predictive cache that prefetches by using an associative memory to recognize access patterns over time [13] Knowledge based caching has been proposed to enhance cache performance of remote file servers [9]. Our work uses a trained neural net to recognize higher level file access patterns from simple reads and writes; like other approaches based on learning algorithms it is extensible and can be trained to include more patterns as the need arises. 9 Conclusions Through automatic input output ....

Korner, K. Intelligent Caching for Remote File Service. In Proceedings of the 10th International Conference on Distributed Computing Systems (May 1990), pp. 220--226.

No context found.

Kim Korner. Intelligent caching for remote file service. In Proceedings of the Tenth International Conference on Distributed Computer Systems, pages 220--226. IEEE Society Press, 1990.

No context found.

Kim Korner. Intelligent Caching for Remote File Service. In Proceedings of the 10th International Conference on Distributed Systems, pages 220--226, May 1990.

No context found.

K. Korner. Intelligent Caching for Remote File Service. In Proc. Tenth Intl. Conf. on Distributed Computing Systems, pages 220--226. IEEE, May 1990.

First 50 documents

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