S. Irani, "Page Replacement with Multi-Size Pages and Applications to Web Caching," Proc. 29th Ann. ACM Symp. Theory of Computing (STOC '97), pp. 701-710, 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of our ongoing work in Section V. II. RELATED WORK File size has long been used as a metric in a wide range of policies and algorithms used to efficiently deliver WWW content. Irani was first to observe that caching policies which do not take file size into account do not work well in practice [9]. Modern streaming media caching algorithms such as Resource Based Caching consider both the impact of file size and required delivery bandwidth in making cache insertion and replacement decisions [15] Size based scheduling policies at Web servers use the size of the requested file to prioritize ....

S. Irani. Page Replacement with Multi-size Pages and Applications to Web Caching. In Proceedings of the 29th Annual ACM Symposium on Theory of Computing, May 1997.

....not only its next occurrence, but also the amount of area that the task will free up upon removal from the chip. This problem has direct applications in web caching where pages have different sizes and request frequencies. A complete discussion along with effective algorithms can be found in [15]. Moreover, although we suppose to have the complete information about the application DFG, many real life applications do not conform to this assumption (the online version of the problems) For instance, a web caching policy has almost no information about the next page that a user might ....

S. Irani. "Page Replacement with Multi-Size Pages and Applications to Web Caching". Algorithmica, 33(3):384--409, 2002.

....marking algorithms. Deterministic Marking Algorithms Randomized Marking Algorithms Type 1 2k 1 2(2H k 1) Type 2 2k 1 4H k Type 3 k k Table 1: Upper bounds on the competitive ratios of all deterministic or randomized marking algorithms. Recent related work. As in the case of a single cache [15, 28, 29], the connection caching model can be generalized by allowing connections to have varying sizes and fetching costs. In the HTTP over TCP context, sizes may correspond to number of socket bu ers allocated. Establishment costs may capture the user perceived latency caused by ....

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proc. 29th Annual ACM Symposium on Theory of Computing. ACM, 1997.

....problems The plain LRU, LFU, and LRFU algorithms are designed for pages with equal sizes and miss costs. Web objects, however, vary significantly in both size and cost of a miss. This motivated the development of cache replacement algorithms that account for varying page sizes and fetching costs [5, 8, 3, 4]. Experiments showed that the optimally competitive Landlord algorithm performs well on Web caching sequences [8, 3] Other experiments, however, show that it can be outperformed by perfect LFU [2] A natural open problem is thus to extend our results to this more general model. That is, develop ....

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proc. 29th Annual ACM Symposium on Theory of Computing. ACM, 1997.

....web environment exhibit di erent characteristics than in traditional operating systems caches. One well recognized di erence is the variability in sizes and fetching costs of di erent resources (pages) and indeed, replacement policies that incorporate these parameters were suggested and evaluated [7, 15, 23, 22]. As is the case in traditional paging, individual requests remain hard to predict. We focus on two separate predictable aspects of the trac in the web environment. The rst is the usage levels experienced by large proxy and (typically) web servers. The second is distinct per page characteristic ....

....next request time of the requested page is provided. When page sizes and fetching costs are uniform, this information suces to perform optimally, by simply evicting the page to be requested furthest in the future [4] However, when fetching costs and sizes vary, this information is insucient [15]. Intuitively, with uniform costs there is a total order on the value of all cached pages (according to their next request time) With varying costs, the relative value of cached pages depends on future requests for pages not yet seen. As a simple example, suppose that the replacement algorithm ....

[Article contains additional citation context not shown here]

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proc. 29th Annual ACM Symposium on Theory of Computing. ACM, 1997.

....time the cache contains a set of pages of total size no greater than k. When a page is requested, if it is cached then the request is a hit. Otherwise it is a miss and the caching algorithm may evict a subset of pages in order to make room for caching the currently requested page. Similar to Irani [7], we charge the algorithm by the respective fetching cost f(p) upon each decision to evict or not to cache a page p. This cost measure differs by an additive constant (incurred by all algorithms) from charging for the fetching cost upon each miss. The goal of a caching algorithm is to minimize ....

....does not necessarily force p into the cache. Forcing a page into the cache is a realistic model of operating system caches but the actual scenario in most web caches is that the page does not have to enter the cache in order to serve the request. This assumption is consistent with that of Irani [7] and is the only detail in which our variant differs from the one considered by Young [15] We prove the following: Theorem 1 The cost of greedy dual size is at most times the cost of the optimal strategy that uses a cache of size h for every h k. We prove this theorem in the next section. ....

[Article contains additional citation context not shown here]

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proc. 29th Annual ACM Symposium on Theory of Computing. ACM, 1997.

....that may occur when requests and their responses are routed within the network. We also remark that cost models have been adopted in uniprocessor caching systems to model scenarios in which the costs incurred in the retrieval of objects on cache misses may vary from one object to another [10, 19, 33]. With regard to uniprocessor caching schemes, recent research has addressed the challenge of designing cache replacement policies that take into account the di ering costs incurred in the retrieval of objects on cache misses. This has led to studies formulating generalizations of the traditional ....

....addressed the challenge of designing cache replacement policies that take into account the di ering costs incurred in the retrieval of objects on cache misses. This has led to studies formulating generalizations of the traditional uniprocessor caching problems that account for the di ering costs [10, 19, 33]. In a recent experimental study [21] Korupolu and Dahlin evaluate the practical performance of several placement and replacement algorithms for cooperative caching. Their simulation experiments demonstrate that, in practice, both our greedy placement algorithm as well as our amortizing ....

S. Irani. Page replacement with multi-size pages and applications to Web caching. In Proceedings of the 29th Annual ACM Symposium on Theory of Computing, pages 701-710, May 1997.

.... (see comment below) Irani considered two special cases of le caching: when the costs are either all equal (the goal is to minimize the number of retrievals) and when each cost equals the le size (the goal is to minimize the total number of bytes retrieved) For these two cases, Irani [7] gave O(log 2 k) competitive randomized on line algorithms. Comment: the importance of sizes and costs. File caching is important for world wide web applications. For instance, in browsers and proxy servers remote les are cached locally to avoid remote retrieval. In web servers, disk les are ....

....the importance of sizes and costs. File caching is important for world wide web applications. For instance, in browsers and proxy servers remote les are cached locally to avoid remote retrieval. In web servers, disk les are cached in fast memory to speed response time. As Irani points out (see [7] and references therein) le size is an important consideration; caching policies adapted from memory management applications that don t take size into account do not work well in practice. Allowing arbitrary costs is likely to be important as well. In many cases, the cost (e.g. latency, total ....

Sandy Irani. Page replacement with multi-size pages and applications to Web caching. Proceedings of the Twenty-Ninth Annual ACM Symposium on Theory of Computing, pages 701-710, El Paso, Texas, 4-6 May 1997.

.... these methodologies and have evaluated several designs in comparison with conventional cache management algorithms (Williams et al. 1996) It is sometimes possible to establish tight bounds on the performance of cache replacement policies independent of workload (client document request patterns (Irani 1997; Sleator and Tarjan 1985) This approach, however, yields performance guarantees far weaker and more pessimistic than the observed performance of most reasonable removal policies under real workloads. In order to evaluate the relative performance of practical caching schemes we must therefore ....

.... documents that are no longer requested (Kelly, Jamin, and MacKie Mason 1999) Extensive experiments have shown that weighted LFU with aging usually delivers more aggregate value to servers than best of breed algorithms from the Web caching literature, e.g. Greedy Dual Size (Cao and Irani 1997). Figure 2: In this trial, weighted LFU (wLFU) provided higher value than unweighted frequency or recency based policies (uLFU and pLRU) Whereas weighted LFU successfully exploits server valuation information to deliver high value to system users, it is not likely to obtain this private ....

Irani, Sandy. 1997. Page replacement with multi-size pages and applications to Web caching. In Twenty-Ninth ACM Symposium on the Theory of Computing, El Paso, TX.

....policies is important. Whereas, models of processor memory hierarchies typically assume mandatory placement (e.g. Sleator and Tarjan on paging policies [34] in Web caching we need not require that a requested document always be cached (as in Irani s discussion of variable page size caching [23]) Optional placement Perfect LFU is optimal for infinite sequences of independent references from a fixed distribution, if document sizes are uniform. Limited empirical evidence, however, suggests that optionalplacement variants of LFU perform worse than their mandatory placement counterparts on ....

S. Irani, Page replacement with multi-size pages and applications to Web caching, 29th ACM STOC, May 1997, pp. 701--710.

....is propagated. Otherwise, the communication cost is zero. Using Definition 3. 9, we have the following definition for meta update propagation: PCC tot(t) X u2U(t) PCC(u; t) 6) Note that our use of the size of a file for calculating the communication cost is identical to the work of Irani [12]. One could have used the size of deltas [15] for calculating the communication cost. We don t do so because it is not clear to us how easy it would be to maintain multiple versions of the documents to account for the different states at the different search engines. The cost function (Cost) ....

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proceedings of the ACM Symposium on Theory of Computing, pages 701--710, 1997.

....(CC) Consider an update u to a le f at time t. Then CC(u) at a later time t 1 is CC(u) size f (w f ) 1) provided that u is propagated before t 1 . Note that our use of the size of a le (or the size of the delta) for calculating the communication cost is identical to the work of Irani [Ira97] 4.3 The cost function Finally, we are ready to provide the cost function. De nition 4.7 Cost: Given an update u to a le f at time t, its cost at a later time t 1 is given by Cost(u) CC(u) OC(u) 2) using the contexts of De nitions 4.3 and 4.6) 5 A competitive algorithm The algorithm ....

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proceedings of the ACM Symposium on Theory of Computing, pages 701-710, 1997.

....p, let size(p) be the size and cost(p) be the cost of p. The total size of the pages in fast memory may never exceed K. The goal is to serve a sequence of requests so that the total loading cost is as small as possible. Various cost models have been proposed in the literature. 1. The Bit Model [38]: For each page p, we have cost(p) size(p) The delay in bringing the page into fast memory depends only upon its size. 2. The Fault Model [38] For each page p, we have cost(p) 1 while the sizes can be arbitrary. 3. The Cost Model: For each page p, we have size(p) 1 while the costs can ....

....of requests so that the total loading cost is as small as possible. Various cost models have been proposed in the literature. 1. The Bit Model [38] For each page p, we have cost(p) size(p) The delay in bringing the page into fast memory depends only upon its size. 2. The Fault Model [38]: For each page p, we have cost(p) 1 while the sizes can be arbitrary. 3. The Cost Model: For each page p, we have size(p) 1 while the costs can be arbitrary. 4. The General Model: For each page p, both the cost and size can be arbitrary. In the Bit Model, and hence in the General Model, ....

S. Irani. Page replacement with multi-size pages and applications to Web caching. Proc. 29th Annual ACM Symp. on Theory of Computing, 701--710, 1997.

....all algorithms. We explore variants of LFU in which reference counts persist across evictions ( Perfect LFU in the terminology of Breslau et al. 6] and in which they are defined only for cached items ( incache LFU ) While some theoretical investigations consider the case of optional placement [13], we find empirically that it never confers a substantial advantage over mandatory placement and often incurs a severe performance penalty, possibly because it ensures that a large fraction of the many twice requested documents in our traces never result in cache hits. Therefore we consider only ....

Sandy Irani. Page replacement with multi-size pages and applications to Web caching. In 29th ACM STOC, pages 701--710, May 1997.

No context found.

S. Irani, "Page Replacement with Multi-Size Pages and Applications to Web Caching," Proc. 29th Ann. ACM Symp. Theory of Computing (STOC '97), pp. 701-710, 1997.

No context found.

S. Irani. Page replacement with multi-size pages and applications to web caching. In 29th ACM Symposium on Theory of Computing, pages 701--710, May 1997.

No context found.

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proc. 29th Annual ACM Symposium on Theory of Computing. ACM, 1997.

No context found.

S. Irani. "Page Replacement with Multi-Size Pages and Applications to Web Caching". Algorithmica, 33(3):384--409, 2002.

No context found.

S. Irani. Page replacement with multi-size pages and applications to web caching, Proc. 29th ACM Symposium on Theory of Computing pp 701-710, 1997.

No context found.

S. Irani. Page replacement with multi-size pages and applications to web caching. In 29th ACM Symposium on Theory of Computing, pages 701--710, May 1997.

No context found.

S. Irani. Page replacement with multi-size pages and applications to Web caching. In Proceedings of the 29th Annual ACM Symposium on Theory of Computing, pages 701--710, May 1997.

No context found.

S. Irani. Page replacement with multi-size pages and applications to web caching. In 29th ACM Symposium on Theory of Computing, pages 701--710, May 1997.

No context found.

S. Irani. Page replacement with multi-size pages and applications to web caching. In Proc. 29th Annual ACM Symposium on Theory of Computing. ACM, 1997.

No context found.

S. Irani. "Page Replacement with Multi-Size Pages and Applications to Web Caching". Algorithmica, 33(3):384--409, 2002.

No context found.

S. Irani. Page replacement with multi-size pages and applications to Web caching. Algorithmica, 33(1):384--409, 2002.

First 50 documents  Next 50

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