S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11:73--91, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....there have been several interesting variants of the framework that were at least partially successful in addressing some of these concerns. Randomized paging algorithms have more realistic performance [5, 11, 13] Some alternative approaches to evaluating on line algorithms were proposed in [1, 12] for the general case and in [2, 6, 7, 15] speci cally for the paging problem. In this paper we propose and study two re nements of competitive analysis which seem to go a long way towards addressing the concerns expressed above. Perhaps more importantly, we show that these ideas give rise to ....

S. Ben-David and A. Borodin, A new measure for the study of on-line algorithms., Algorithmica, 11 (1994), pp. 73-91.

....First Fit has a strictly better accommodating ratio than Worst Fit. AMS subject classification: 68Q25, 05B40, 90B35 1 Introduction The competitive ratio [13, 24, 19] as a measure for the quality of on line algorithms, has been criticized for giving bounds that are unrealistically pessimistic [1, 3, 14, 18, 21], and for not being able to distinguish between algorithms with very different behavior in practical applications [3, 14, 21, 23] Though this criticism also applies to standard worst case analysis, it is often more disturbing in the on line scenario [14] The basic problem is that the adversary ....

S. Ben-David and A. Borodin. A New Measure for the Study of On-Line Algorithms. Algorithmica, 11:73--91, 1994.

....each request is repeated l times, thus, hiding the future requests from the algorithm. Empirically, however, on line algorithms benefit from using such a lookahead and several authors have suggested other models for the evaluation of on line algorithms in which this lookahead makes a difference [3, 7, 6, 10]. Young [11] suggested an alternative measure for the lookahead size which he calls resource bounded lookahead. He showed that there exists an on line paging algorithm with resource bounded lookahead l that has competitive ratio 1 max n 2 Delta K k k l 1 ; 2 o , and that no such ....

S. Ben-David and A. Borodin. A New Measure for the Study of On-Line Algorithms. Algorithmica, 11:73--91, 1994.

....MIN. They showed that LRU s performance penalty can be proportional to the size of fast memory, but no other on line algorithm can, in the worst case, do much better. These fundamental results have been extended in several ways, most often to include models that allow different forms of lookahead [3, 1, 9, 7]. All these works deal with the question of which buffer block to evict. In contrast, in our situation the additional question that arises is when to fetch a block and evict some other. Cao et al. 4] examined prefetching from a single disk to overlap CPU and I O operations. They defined two ....

S. Ben-David and A. Borodin. A New Measure for the Study of On-Line Algorithms. Algorithmica, 11:73--91, 1994.

....accessed farthest in the future. On line buffer management was studied in [42] where the competitive ratio of on line page replacement policies, LRU and FIFO were shown to be (M) where M is the size of the I O buffer. Extended models incorporating lookahead (and extra buffers) were studied in [3, 8, 1, 11, 22, 31, 44] in the sequential I O case. The problem there is to determine which blocks need to be replaced in an on line manner, with the measure being the competitive ratio. It has been shown that the on line page replacement strategies like LRU and FIFO belong to a general class of marking algorithms which ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11(1):73--91, January 1994.

....a specialist and an online algorithm that adapts well to all hint values is called a generalist. A specialist is similar to a champion in the measure of competitive latency. Other measures for studying the performance of online algorithms include the Max Max measure by Ben David and Borodin [5] and the Competitive Throughput measure by Aspnes and Waarts [2] In the definition of the stability ratio, we model the probability of local reads at node u by pru and the probability of local writes at node u by pwu , such that X u (pru pwu) 1. A global history of the system is a ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 4:73--91, 1994.

....can look into the future up to some total cost, rather than a number of items into the future. While these papers allow the on line algorithms to have some knowledge about the future, the problems themselves are very different from the acknowledgment delay problem. Finally, Ben David and Borodin [2] studied the problem of look ahead for the k server problem. They showed that any finite look ahead L could not improve the competitive ratio by simply repeating each request in the request sequence L times. While we obtain the same type of result for the our problem, a more involved lower bound ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11:73--91, 1994.

....there have been several interesting variants of the framework that were at least partially successful in addressing some of these concerns. Randomized paging algorithms have more realistic performance [5, 11, 15] Some alternative approaches to evaluating on line algorithms were proposed in [1, 14] for the general case and in [2, 6, 7, 17] specifically for the paging problem. In this paper we propose and study two refinements of competitive analysis which seem to go a long way towards addressing the concerns expressed above. Perhaps more importantly, we show that these ideas give rise to ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11(1):73--91, January 1994.

....will mention related work relevant to the specific problems we consider. Here, we o#er a brief overview of generally relevant related work. The importance of when information becomes available has been noted previously, especially in the significant body of work on algorithms with lookahead, e.g. [12, 22, 24, 28]. In the case of on line decision models, however, the possibility of not having up to date information is not generally addressed. For load balancing problems, the question has been considered for statistical models [30, 31, 37] And recently, 5] considers an on line load balancing setting where ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11:73-- 91, 1994.

....will mention related work relevant to the specific problems we consider. Here, we offer a brief overview of generally relevant related work. The importance of when information becomes available has been noted previously, especially in the significant body of work on algorithms with lookahead, e.g. [12, 20, 22, 25]. In the case of on line decision models, however, the possibility of not having up to date information is not generally addressed. For load balancing problems, the question has been considered for statistical models [27, 28, 34] other queueing based models have also been considered [4, 5, 26] ....

S. Ben-David and A. Borodin. A New Measure for the Study of On-Line Algorithms. Algorithmica, 11:73--91, 1994.

....wait for some incoming requests. In the theoretical context a natural question is: What is it worth to know a part of the future So far, only few on line problems with lookahead have been studied in the literature. Previous research on lookahead in on line algorithms has addressed paging problems [1, 5, 11, 23, 29, 30], k server problems [5] bin packing problems [16] dynamic location problems [14] and graph problems [17, 19, 21] In particular, at present time, nothing is known about list update with lookahead. We begin our study of the influence of lookahead in the list update problem by introducing two ....

....context a natural question is: What is it worth to know a part of the future So far, only few on line problems with lookahead have been studied in the literature. Previous research on lookahead in on line algorithms has addressed paging problems [1, 5, 11, 23, 29, 30] k server problems [5], bin packing problems [16] dynamic location problems [14] and graph problems [17, 19, 21] In particular, at present time, nothing is known about list update with lookahead. We begin our study of the influence of lookahead in the list update problem by introducing two different models of ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11(1):73-91, 1994.

....logarithmically in price variance. 1. 2 Definitions and simple examples In addition to the standard competitive ratio there were some alternative approaches or refinements to evaluating on line algorithms and distributed decisions making with incomplete information such as [AP91, BFR92, AKP92, BB94, BIRS91, XP92, PY93, KP94, AADW94] However, none of the above capture the issue of incomplete information for centralized algorithm. To see that the standard competitive analysis setting does not capture above scenario, it is sufficient to note that, in case of economic warfare, an online ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11(1):73--91, 1994.

....for paging strategies are too high to be of practical interest. Simulations show that, in practice, good paging strategies usually incur a cost within a small constant factor of the minimum [65] Another criticism was that competitive analysis can be counterintuitive. Ben David and Borodin [13] gave the following example, when we choose to abandon all competitive strategies in favor of a noncompetitive one. Consider buying an insurance policy. Paying 5 a month to insure the car is a noncompetitive strategy, because there is a scenario where one will never present a claim to the ....

....in this model, the competitive ratio of LRU is less than or equal to the competitiveness ratio of FIFO, a behavior often observed in practice. Chrobak and Noga [25] later proved this conjecture. Another model, which departs from the competitiveness model, was proposed by Ben David and Borodin [13]. Their Max Max ratio compares the worst case amortized behavior of an algorithm with that of an optimal offline algorithm. This is a revival of the worst case amortized analysis, except that now it is normalized by the best that can be done using an optimal offline algorithm. They show that this ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11(1):73--91, 1994.

....Science, University of Southern Denmark, Main Campus: Odense University, Campusvej 55, DK 5230 Odense M, Denmark. 1 Introduction The competitive ratio [14, 26, 19, 23] as a measure for the quality of online algorithms, has been criticized for giving bounds that are unrealistically pessimistic [4, 5, 15, 18, 20], and for not being able to distinguish between algorithms with very different behavior in practical applications [5, 15, 20, 24] Though this criticism also applies to standard worst case analysis, it is often more disturbing in the on line scenario [15] The basic problem is that the adversary ....

S. Ben-David and A. Borodin. A New Measure for the Study of On-Line Algorithms. Algorithmica, 11:73--91, 1994.

....Odense University, Campusvej 55, DK 5230 Odense M, Denmark. This decision must be based on some measure (and analysis) So in order for the decision to be made correctly, measures (and analyses) must be accurate. The competitive ratio has been criticized for being unrealistically pessimistic [1, 2, 10, 12, 14], and for not being able to distinguish between algorithms with very different behavior in practical applications [2, 10, 14, 15] One reason for this is that a single very long and very unusual sequence can bring the ratio up to a level very far from the result obtained on typical sequences. This ....

S. Ben-David and A. Borodin. A New Measure for the Study of On-Line Algorithms. Algorithmica, 11:73--91, 1994.

.... fact, some paging algorithms used in practice make use of lookahead [15] In the theoretical context a natural question is: What is it worth to know a part of the future Previous research on lookahead in on line algorithms has mostly addressed dynamic location problems and on line graph problems [4, 9, 7, 6, 8]; only very little is known in the area of competitive paging with lookahead. Consider the intuitive model of lookahead, which we call weak lookahead. Let l 1 be an integer. We say that an on line paging algorithm has a weak lookahead of size l if it sees the present request to be served and the ....

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11(1):73--91, 1994.

....the limitations of competitive analysis. These methods include restricting the input distribution to satisfy some special properties (e.g. 10, 16] and comparing the cost of a solution produced by an online algorithm on input I to the worst case optimal cost of any input of the same size as I [5]. 3 Average Response Time The following well known lemma explains why we first consider the queue size. Lemma 3.1 For any scheduling algorithm A with a speed s processor, the total response time is R 1 0 jQA (t; s)jdt. Lemma 3.2 For every nonclairvoyant scheduling algorithm A there is an ....

S. Ben-David, and A. Borodin, "A new measure for the study of on-line algorithms", Algorithmica, 11, 73--91, 1994.

No context found.

S. Ben-David and A. Borodin. A new measure for the study of on-line algorithms. Algorithmica, 11:73--91, 1994.

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