E. Cohen, B. Krishnamurthy, and J. Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of the 6th European Symposium on Algorithms, pages 307--319. Springer-Verlag, Lecture Notes in Computer Science Vol. 1461, August 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....a better model of what is likely to be requested next. Therefore, in a client based prefetching system, the predictions need to be sent to the client in some form. A number of researchers have proposed the use of hints of some kind for caching, invalidation, or prefetching [Pad95, PM96, Mog96, CKR98, KW97, KW98b, Duc99] While the hints themselves could be delivered through HTTP using a particular unique URL, the client still needs to know the identity of the appropriate URL that would contain hints of actions likely to follow a particular object retrieval. One possibility is to calculate ....

Edith Cohen, Balachander Krishnamurthy, and Jennifer Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of the 6th European Symposium on Algorithms, August 1998. LLNCS vol. 1461.

....Cross domain work is highlighted with italicized text. The examples that will be presented in later chapters are explicitly named in the table. Prediction Generation Server Proxy Client External [Gwe95] Wes95] Bes96] Server [BC96] MC96] CV971 [NB97] LG96] DCW96] Proxy Push [TDV98] CKR98] DCW96] CY97] Proxy Pull [JC98] YZL01] PM96] FCL99] PM94] E mail Veb [HWM981 [JK971 Client [Duc99] cache mgmt [croo] Veb DNS Table 4.1: Web Prediction Generation and Execution 4.2 Predictive Information This section explores the information that is used to make predictions. Chapter 2 ....

Cohen, E., Krishnamurthy, B., and Rexford, J., "Evaluating Server-Assisted Cache Replacement in the Web", in Proceedings of the Sixth European Symposium on Algorithms, Springer-Verlag, Lecture Notes in Computer Science, August 1998, v. 1461, pp. 307-319.

....of different web servers on a proxy server. Instead, semantic hints(such as path information) are disseminated from web servers to proxy servers to facilitate their page replacement decisions. Server hinted caching or prefetching has been studied in the literature and has been proven effective [19, 37, 40]. Generally, a web server, on which a prediction model resides, sends knowledge of access patterns to proxies by a specific protocol, or by modified HTTP. Proxy servers utilize this knowledge to optimize their caching replacement strategies or prefetching policies(Figure 4.1) In our later ....

E. Cohen, B. Krishnamurthy, and J. Rexford. Evaluating server-assisted cache replacement in the web. In Proceedings of European Symposium on Algorithms, August 1998.

....its rules and returns predictions as hints to proxy servers. The proxy server then utilizes this information to determine its caching or prefetching strategy. Server hinted architecture has been studied extensively and proven effective in the context of model based Web caching and prefetching [14, 15]. In our later discussion, we assume that proxy servers can receive all the hints from the web servers on an ad hoc basis. We also assume that the network Figure 2 Prediction Algorithm Input: W: access sequence of current session; h c : confidence threshold; T: prediction rule table; EOT of T ....

E. Cohen, B. Krishnamurthy, and J. Rexford. Evaluating server-assisted cache replacement in the web. In

....streams. Exploiting Temporal Locality Properties: Locality of reference properties have been exploited in a number of Web caching, replication, and prefetching protocols and systems. Generally speaking, such protocols can be classi ed as: 1) Server based (i.e. at the server or a proxy thereof) [7, 8, 15], 2) Clientbased (i.e. at the client or a proxy thereof) 1, 9] or (3) Network based (i.e. in the network, transparent to both the server and the client) 23, 12, 28, 14, 33, 20] An important reason for this classi cation is that Web reference characteristics are likely to be di erent in each ....

Edith Cohen, Balachander Krishnamurthy, and Jennifer Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of ESA'98, 1998.

....Exploiting Temporal Locality Properties: Locality of reference properties have been exploited in a number of Web caching, replication, and prefetching protocols and systems. Generally speaking, such protocols can be classi ed as: 1) Server based (i.e. at the server or a proxy thereof) [7, 8, 13], 2) Client based (i.e. at the client or a proxy thereof) 1, 9] or (3) Network based (i.e. in the network, transparent to both the server and the client) 18, 11, 27] An important reason for this classi cation is that Web reference characteristics are likely to be di erent in each of the above ....

Edith Cohen, Balachander Krishnamurthy, and Jennifer Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of ESA'98, 1998.

....time cost and the size. # Bolot Hoschka [10] employs a weighted rational function of transfer time cost, the size, and the time last access. # Size Adjusted LRU (SLRU) 3] orders the object by ratio of cost to size and choose objects with the best costto size ratio. # Server assisted scheme [19] models the value of caching an object in terms of its fetching cost, size, next request time, and cache prices during the time period between requests. It evicts the object of the least value. # Hierarchical GreedyDual (Hierarchical GD) 40] does object placement and replacement cooperatively in ....

E. Cohen, B. Krishnamurthy, and J. Rexford, Evaluating server-assisted cache replacement in the Web, Proceedings of the European Symposium on Algorithms-98, 1998.

....properties, especially if coupled with server file caching strategies. Our current arbitration algorithms considers all clients as equal. It has been shown that performance improvement in the context of server aided caching can be obtained by considering a small number of di#erent client classes [11]. Di#erent and more e#cient cache replacement strategies can stem from a tighter interaction between servers and clients. We have only considered dissemination of non dynamic documents, which form the bulk of our traces, but several issues would arise if dynamic contents were a more significant ....

Edith Cohen, Balachander Krishnamurthy, and Jennifer Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of the European Symposium on Algorithms, 1998.

....eviction decisions are determined based on these distributions. This framework was previously used in various contexts (for example, holding times for IP over ATM virtual circuits [25] spinning on a lock in a shared memory multiprocessor [24] adaptive disk spindown [27] and Web proxy caching [12]. Theoretical analysis and polices were given in [10, 24, 27, 28] an optimal policy in the holding time model was given in [24, 25, 10] and an optimal policy in the caching model, for predictable load conditions, was provided in [10] The optimal policy in this context uses knowledge of the ....

....performs well in practice. Classical theoretical results establish that lru is k k Gammah 1 competitive against the optimal offline algorithm with cache of size h k [33] The policy analogous to lru in the holding time model is the one that uses the same interval T (r) for every request r [10, 25, 12]. Hereafter we use the name lru for this policy. lru is parameterized by the constant V to which T (r) is set. 5.2 opt As a benchmark we consider the performance of an omniscient optimal policy. The optimal policy knows the sequence in advance and achieves the best trade offs of misses and ....

E. Cohen, B. Krishnamurthy, and J. Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of the 6th European Symposium on Algorithms, pages 307--319. Springer-Verlag, Lecture Notes in Computer Science Vol. 1461, August 1998.

....on the time until the next request for the requested page. This model was rst considered by Lund et al. 19] in the context of holding time policies for virtual circuits (for routing IP trac over ATM [17, 18] and was termed IP paging. The same model was proposed later on for web caching [20, 10]. Following Lund et al. 19] our yard stick is the optimal online algorithm which makes best use of knowing the underlying probability space, but does not know the actual sequence drawn from it apriori. This algorithm is well de ned by Markov decision theory for discrete distributions, but is ....

....interval caching algorithm. Simulating the interval caching algorithm is possible using only a partial trace, and in case the load on the cache is indeed steady, both simulations should produce similar results. One example of such experiment is the evaluation of server assisted cache replacement [10]. In server assisted cache replacement, the web server collects statistics on access patterns to its pages, and disseminates it as hints to its clients that enhance cache with uniform fetching costs and sizes. with respect to the miss rate. The median algorithm does not have bounded ....

[Article contains additional citation context not shown here]

E. Cohen, B. Krishnamurthy, and J. Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of the 6th European Symposium on Algorithms, pages 307-319. Springer-Verlag, Lecture Notes in Computer Science Vol. 1461, August 1998.

....both content and freshness misses. The content hit rate is measured per object or per byte and sometimes weighted by estimated object fetching cost. It is dictated by the available cache storage and the replacement policy used. Replacement policies for Web caches were extensively studied (e.g. [3, 4, 5, 6, 7, 8, 9, 10]) Policies that seem to perform well are Least Recently Used (lru, which evicts the least recently requested object when the cache is full) Least Frequently Used (lfu, which evicts the least frequently requested object) and Greedy Dual Size (which accounts for varying object sizes and fetching ....

E. Cohen, B. Krishnamurthy, and J. Rexford, \Evaluating server-assisted cache replacement in the Web," in Proceedings of the 6th European Symposium on Algorithms, pp. 307-319, Springer-Verlag, Lecture Notes in Computer Science Vol. 1461, Aug. 1998.

....and freshness misses. The content hit rate is measured per object or per byte and sometimes weighted by estimated object fetching cost. It is dictated by the available cache storage and the replacement policy used. Replacement policies for Web caches were extensively studied (e.g. 2] 3] 4] [5], 6] 7] 8] 9] Policies that seem to perform well are Least Recently Used (LRU, which evicts the least recently requested object when the cache is full) Least Frequently Used (LFU, which evicts the least frequently requested object) and GreedyDual Size (which accounts for varying object ....

E. Cohen, B. Krishnamurthy, and J. Rexford, "Evaluating server-assisted cache replacement in the Web," in Proceedings of the 6th European Symposium on Algorithms, pp. 307--319, Springer-Verlag, Lecture Notes in Computer Science Vol. 1461, Aug. 1998.

....a new query. The different policies vary by when an entry is renewed. We consider several natural policies, based on reference locality (analogous to the cache replacement algorithm LRU) access frequency (analogous to LFU) and an adaptive perhostname policy (analogous to policies studied in [8, 5, 17]) These analogies are made based on properties of the request sequence that are exploited but the underlying cost benefit measures are different. We experimentally evaluated and compared performance of the different polices using a heterogeneous set of proxy logs, consistently obtaining ....

....Each host h that had sufficiently many misses under R LRU with r = 1 on the learning data, gets a value r(h) 1. A generic value r(h) r 1 is used for hosts for which there was no sufficient data. The mapping of hosts to values r(h) is performed using similar techniques as the ones used in [8, 7] and analyzed in [5] The implementation of R ADAPTIVE is considerably more involved than R LRU, R FIFO, or R LFU and requires generating and maintaining the mapping r(h) R OPT: is the optimal renewal policy. The performance curve of R OPT gives for any number of renewals, the minimum possible ....

E. Cohen, B. Krishnamurthy, and J. Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of the 6th European Symposium on Algorithms, pages 307--319. Springer-Verlag, Lecture Notes in Computer Science Vol. 1461, Aug. 1998.

....on the time till the next request for the requested page. This model 2 was first considered by Lund et al. 17] in the context of holding time policies for virtual circuits (for routing IP traffic over ATM [15] and was termed IP paging. The same model was proposed later on for web caching [18, 9]. Following Lund et al. 17] our yard stick is the optimal online algorithm which makes best use of knowing the underlying probability space, but does not know the actual deterministic sequence apriori. This algorithm is well defined by Markov decision theory for discrete distributions, but is ....

....Evaluation using a standard model necessitates a full workload for the cache. Interval caching evaluation is performed using the analogous policies, and considers trade offs of hit rate and average cache size. One example of such experiment is the evaluation of server assisted cache replacement [9]. In server assisted cache replacement, the web server collects statistics on access patterns to its pages, and disseminates it as hints to its clients that enhance cache replacement in the clients caches. Full workload evaluation of server assisted vs. other replacement policies requires a ....

[Article contains additional citation context not shown here]

E. Cohen, B. Krishnamurthy, and J. Rexford. Evaluating server-assisted cache replacement in the Web. In Proceedings of the 6th European Symposium on Algorithms, pages 307-- 319. Springer-Verlag, Lecture Notes in Computer Science Vol. 1461, August 1998.

....clients, is in a unique position to gauge the popularity of its resources, and the likelihood that certain resources are accessed together. For example, the server can speculatively disseminate resources to the proxy [1] guide proxy prefetching decisions [2 5] guide cache replacement decisions [6, 7], and improve cache coherency by sending information about modified resources [8, 9] The overhead can be minimized by focusing on resources that are likely to be requested in the future. In earlier work, we introduced an end to end framework for server response messages to include hints (in ....

E. Cohen, B. Krishnamurthy, and J. Rexford, "Evaluating server-assisted cache replacement in the web," in Proc. European Symposium on Algorithms, August 1998.

....near future. Recent studies have begun to consider the potential benefits of using server information to enhance the proxy s policies. For example, the server can aid the proxy in cache allocation and replacement decisions by sending estimates of the time between successive accesses to a resource [13, 24]. Similarly, using access patterns to predict future requests, the server can speculatively disseminate resources to the proxy [14] or guide proxy prefetching decisions [15 18] at the expense of increasing the load on the network and the server. In addition, the server can improve cache ....

....metric shows that the piggyback messages generate a considerable number of predictions for requests for cached resources. In addition, these predictions are accurate. Drawing on these promising initial results, we have performed an extensive comparison of server assisted cache replacement policies [24]. Adaptive freshness interval: The proxy can use the piggyback message to tune its caching policies, even when the piggybacked items do not reside in the cache. Since the piggyback includes the Last Modified time of each resource, the proxy can estimate and record how often the resource changes, ....

E. Cohen, B. Krishnamurthy, and J. Rexford, "Evaluating server-assisted cache replacement in the web," in Proc. European Symposium on Algorithms, August 1998. http://www.research.att.com/~bala/papers/ esa98.ps.gz

....decisions [2 5] improve cache coherency by sending information about modified resources [6, 7] The overhead can be minimized by focusing on the resources that are most likely to be requested in the future. Proxy s cache replacement policy can be enhanced with estimates of future accesses [8, 9] . In earlier work, we introduced an end to end framework for server response messages to include hints (in groups called volumes) about future requests, customized to the requesting proxy [10] Here, we present a detailed design and evaluation of algorithms for grouping resources that are ....

E. Cohen, B. Krishnamurthy, and J. Rexford, "Evaluating server-assisted cache replacement in the web," in Proc. European Symposium on Algorithms, August 1998. http://www.research.att.com/~bala/papers/esa98.ps.gz.

....decisions [2 5] improve cache coherency by sending information about modified resources [6, 7] The overhead can be minimized by focusing on the resources that are most likely to be requested in the future. Proxy s cache replacement policy can be enhanced with estimates of future accesses [8, 9]. In earlier work, we introduced an end to end framework for server response messages to include hints (in groups called volumes) about future requests, customized to the requesting proxy [10] Here, we present a detailed design and evaluation of algorithms for grouping resources that are likely ....

E. Cohen, B. Krishnamurthy, and J. Rexford, "Evaluating server-assisted cache replacement in the web," in Proc. European Symposium on Algorithms, August 1998. http://www.research.att.com/~bala/papers/esa98.ps.gz.

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