S. Jin and A. Bestavros. GreedyDual* Web Caching Algorithm: Exploiting the Two Sources of Temporal Locality in Web Request Streams. International Journal on Computer Communications, 24(2):174--183, February 2001.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....major concern. Enormous research efforts have also been put into characterizing the Web [7, 21] and file system [86] workloads and many static cache replacement policies have been invented. Today, robust static policies that work well with a wide variety of workloads are embedded into the systems [59, 20, 84]. Unfortunately, these policies cannot adapt to changes in workload and network topology and become suboptimal [99] when the conditions change. Many factors increase the complexity of today s systems [15] in which caching is used. First, the characteristics of workloads change over short and long ....

....and object size are the most commonly used criteria for local replacement decisions. Least Recently Used (LRU) uses recency of access as the sole criteria for replacement, while Least Frequently Used (LFU) uses frequency of access. SIZE replaces the largest object and Greedy Dual Size (GDS) [59, 29] replaces the object with the smallest key K i C i S i L, where C i is the retrieval cost, S i is the size and L is a running age factor. GDS with Frequency (GDSF) 20] adds the frequency of access, F i , into the same equation and replaces the object with the smallest key K i S i ....

[Article contains additional citation context not shown here]

S. Jin and A. Bestavros. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

....objects which it thinks will cause the fewest or least expensive future misses. In this work we consider twelve baseline policies including seven common policies (RAND, FIFO, LIFO, LRU, MRU, LFU, and MFU) and five more recently developed and very successful policies (SIZE and GDS [CI97] GD [JB00] GDSF and LFUDA [ACD 99] These algorithms employ a variety of directly observable criteria including recency of access, frequency of access, size of the objects, cost of fetching the objects from secondary memory, and various combinations of these. The primary difficulty in selecting the ....

Shudong Jin and Azer Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. Technical Report 2000-011, 4, 2000.

....thinks will cause the fewest or least expensive future misses. 1. 1 Caching Policies In this work we consider twelve baseline policies including seven common policies (RAND, FIFO, LIFO, LRU, MRU, LFU, and MFU) and five more recently developed and very successful policies (SIZE and GDS [9] GD [17], GDSF and LFUDA [2] These algorithms employ a variety of directly observable criteria including recency of access, frequency of access, size of the objects, cost of fetching the objects from secondary memory, and various combinations of these. Table 1.1 roughly groups these twelve policies by ....

Shudong Jin and Azer Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. Technical Report 2000-011, 4, 2000.

....on supporting a single family of strategies. For example, the TACT toolkit [41] provides support for replication based on anti entropy schemes [10] for a range of consistency models. In a similar fashion, caching algorithms exist that base their decisions on temporal correlations between requests [16, 36], but otherwise essentially follow the same protocol. Closer to our approach are systems that have protocols that adapt the way updates are propagated. For example, the adaptive leases described in [12] provide a way for switching from a protocol in which updates are pushed to replicas, to one in ....

Shudong Jin and Azer Bestavros, GreedyDual* Web Caching Algorithm: Exploiting the two Sources of Temporal Locality in Web Request Streams, Comp. Comm. 24 (2001), no. 2, 174--183.

....objects which it thinks will cause the fewest or least expensive future misses. In this work we consider twelve baseline policies including seven common policies (RAND, FIFO, LIFO, LRU, MRU, LFU, and MFU) and five more recently developed and very successful policies (SIZE and GDS [CI97] GD [JB00] GDSF and LFUDA [ACD 99] These algorithms employ a variety of directly observable criteria including recency of access, frequency of access, size of the objects, cost of fetching the objects from secondary memory, and various combinations of these. The primary difficulty in selecting the ....

Shudong Jin and Azer Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. Technical Report 2000-011, 4, 2000.

....the storage nodes have been filled. We do not consider any variants of these base heuristics, such as using popularity divided by object size, but evaluate only the performance of the base heuris tics. Different variants of these heuristics have been studied in the context of web caching (see [8] and references therein) and any such improvements could be used directly to enhance the performance of our heuris tics. 5 Evaluation of Heuristics We evaluated the performances of our heuristics using the topologies from Section 2.1. We ran each heuristic on each topology using different ....

S. Jin and A. Bestavros. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of 5th Web Caching and Content Distribution Workshop, Lisbon, Portugal, May 22-24, 2000.

....the storage nodes have been filled. We do not consider any variants of these base heuristics, such as using popularity divided by object size, but evaluate only the performance of the base heuris tics. Different variants of these heuristics have been studied in the context of web caching (see [8] and references therein) and any such improvements could be used directly to enhance the performance of our heuris tics. 5 Evaluation of Heuristics We evaluated the performances of our heuristics using the topologies from Section 2.1. We ran each heuristic on each topology using different ....

S. Jin and A. Bestavros. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of 5th Web Caching and Content Distribution Workshop, Lisbon, Portugal, May 22-24, 2000.

....of the applications. Enormous research efforts have been put into characterizing Web [1] and file system [25] workloads, and many static cache replacement policies have been invented. Today, robust static policies that work well with a wide variety of workloads are embedded into systems [19, 5, 24]. Unfortunately, these policies cannot adapt to changes in workload and network topology and become suboptimal when the conditions become more complex than the characterized cases [30] Many factors increase the complexity of today s systems in which caching is used. First, the characteristics of ....

....criteria for replacement, while Least Frequently Used (LFU) uses frequency or popularity of access. Most Recently Used (MRU) and Most Frequently Used (MFU) are not successful when used alone, but may be beneficial in mixtures of policies. SIZE replaces the largest object and Greedy Dual Size (GDS) [19, 8] replaces the object with the smallest key K i C i S i L, where C i is the retrieval cost, S i is the size and L is a running age factor. L is set to the key value of the objects that are replaced from the cache. GDS with Frequency (GDSF) 5] adds the frequency of access, F i , into the ....

[Article contains additional citation context not shown here]

JIN, S., AND BESTAVROS, A. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of the 5th International Web Caching and Content Delivery Workshop (Lisbon, Portugal, May 2000).

....workloads measured in 1998 HTML and image documents account for over 95 of all requests. The optimization of cache replacement schemes is important because the growth rate of web content (i.e. multi media documents) is much higher than anticipated growth of memory sizes for future web caches [8]. Furthermore, recent studies (see e.g. 3] have shown hit rate and byte hit rate grow in a log like fashion as a function of size of the web cache. Cao and Irani introduced the web cache replacement scheme Greedy Dual Size (GDS [4] that takes into account document sizes and a user defined cost ....

....scheme Greedy Dual Size (GDS [4] that takes into account document sizes and a user defined cost function. They proved that GDS is on line optimal with respect to this cost function. Jin and Bestavros introduced the web cache replacement scheme Greedy Dual (GD ) as an improvement to GDS [8]. They compared the performance of this newly proposed replacement scheme with traditional schemes as Least Recently Used (LRU) Least Frequently Used with Dynamic Aging (LFU DA) and with the size aware scheme GDS [8] Arlitt, Friedrich, and Jin provided a comparative performance study of six web ....

[Article contains additional citation context not shown here]

S. Jin and A. Bestavros, Greedy Dual* Web Caching Algorithm: Exploiting the Two Sources of Temporal Locality in Web Request Streams. Computer Communications Special Issue on 5 th Web Caching and Content Delivery Workshop, 22, 174-183, 2000.

....and object size are the most commonly used criteria for local replacement decisions. Least Re2 cently Used (LRU) uses recency of access as the sole criteria for replacement, while Least Frequently Used (LFU) uses frequency of access. SIZE replaces the largest object and Greedy Dual Size (GDS) [14] replaces the smallest key K i C i S i L, whereC i is the retrieval cost, S i is the size and L is a running age factor. GDS with Frequency (GDSF) adds the frequency of access, F i , into the same equation. Lowest Relative Value (LRV) 18] replacement makes a cost benefit analysis ....

....First, the possible criteria and the ways to use them are endless. Second, the trend in cache replacement algorithms is towards finding the functions that unite all the criteria in a single key or value. Other taxonomies of time, frequency and size based policies are presented in prior work [14, 6]. Virtual cache management [2] divides the cache into static partitions and lets a few successful policies work in separate partitions. Objects evicted from one partition go to the next until they are moved out of the cache. This scheme is not as homogeneous as our voting scheme and the ....

S. Jin and A. Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request stream. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

....on supporting a single family of strategies. For example, the TACT toolkit [40] provides support for replication based on anti entropy schemes [9] for a range of consistency models. In a similar fashion, caching algorithms exist that base their decisions on temporal correlations between requests [15, 35], but otherwise essentially follow the same protocol. Closer to our approach are systems that have protocols that adapt the way updates are propagated. For example, the adaptive leases described in [11] provide a way for switching from a protocol in which updates are pushed to replicas, to one in ....

Shudong Jin and Azer Bestavros, GreedyDual* Web Caching Algorithm: Exploiting the two Sources of Temporal Locality in Web Request Streams, Comp. Comm. 24 (2001), no. 2, 174-183.

....is a simple on line estimation of 0 T . Of course, the ORCL strategy can be simulated only on historical data, since we need to know future requests in order to compute 0 T . Many other algorithms can be found in literature, very similar to or extending the schema of the above strategies [2,3,14,20,21,23,28,29,33,36,39,46]. Also, we refer to [5,8] for the state of the art in theoretical complexity analysis of on line and o line caching strategies. However, we notice that all such approaches and re nements are all of a static nature, since they follow some xed criteria usually originated from an a priori analysis ....

....in presence of some form of authentication (e.g. login or cookies) 3. 5 Characterization of workloads used in experiments There is an extensive literature on understanding the statistical characteristics of workloads of web proxy servers [7,9,12,24,30] and their impact on caching policies [7,28,29]. In this section, we introduce two workloads that will be used in experimentation and show that their statistical distributions re ect the typical characteristics reported in the literature, with particular reference 12 0 0.2 0.4 0.6 0.8 1 1 10 100 1000 10000 100000 1e 06 Cumulative ....

S. Jin and A. Bestavros, GreedyDual* Web caching algorithms: Exploiting the two sources of temporal locality in Web request streams, in: Proc. of the Int'l Web Caching and Content Delivery Workshop (2000).

No context found.

S. Jin and A. Bestavros. GreedyDual* Web Caching Algorithm: Exploiting the Two Sources of Temporal Locality in Web Request Streams. International Journal on Computer Communications, 24(2):174--183, February 2001.

No context found.

S. Jin and A. Bestavros. Greedydual* Web caching algorithm: Exploiting the two sources of temporal locality in Web request streams. In Proc. of Web Caching Workshop, May 2000.

No context found.

S. Jin and A. Bestavros. GreedyDual* Web Caching Algorithm: Exploiting the Two Sources of Temporal Locality in Web Request Streams. International Journal on Computer Communications, 24(2):174--183, February 2001.

No context found.

S. Jin and A. Bestavros. Greedydual* Web caching algorithm: Exploiting the two sources of temporal locality in Web request streams. In Proc. of Web Caching Workshop, May 2000.

....fall under two categories: 1) optimizations that improve resource allocation decisions, and (2) optimizations that boost resource utilization. Examples of approaches that aim to improve resource allocation decisions include the use of better cache management [13] 10] 9] 8] 24] [21] or the use of prefetching [16] 27] Examples of approaches that aim to improve resource utilization include the elimination of unnecessary memory transfers between the various layers in the system (user space, kernel space, network buffers, etc. 29] and avoiding overloading through proper ....

S. Jin and A. Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request stream. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

....fall under two categories: 1) optimizations that improve resource allocation decisions, and (2) optimizations that boost resource utilization. Examples of approaches that aim to improve resource allocation decisions include the use of better cache management [13] 10] 9] 8] 24] [21] or the use of prefetching [16] 27] Examples of approaches that aim to improve resource utilization include the elimination of unnecessary memory transfers between the various layers in the system (user space, kernel space, network buffers, etc. 29] and avoiding overloading through proper ....

S. Jin and A. Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request stream. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

No context found.

S. Jin and A. Bestavros. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Int'l Journal on Computer Communications,, Vol.24(2), pp.174--183, Feb. 2001.

No context found.

S. Jin and A. Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

No context found.

Shudong Jin and Azer Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. Technical Report 2000-011, 4, 2000.

No context found.

S. Jin and A. Bestavros. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

No context found.

S. Jin and A. Bestavros. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

No context found.

S. Jin and A. Bestavros. GreedyDual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. In Proceedings of the 5th International Web Caching and Content Delivery Workshop, Lisbon, Portugal, May 2000.

No context found.

Shudong Jin and Azer Bestavros. Greedydual* web caching algorithm: Exploiting the two sources of temporal locality in web request streams. Technical Report 2000-011, 4, 2000.

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