Bala Kalyanasundaram and Kirk Pruhs, "Speed is as powerful as clairvoyance," Journal of the ACM, vol. 47, no. 4, pp. 617--643, 2000.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....should be limited to level the playing field. The most popular approach for doing this is to explicitly limit the adversary in some way, e.g. by limiting its freedom in choosing its inputs. In this paper, we use an alternative approach that has been utilized very successfully in scheduling theory [Edm99, KP00, EP01], viz. giving the online algorithm strictly more resources compared to the adversary. Using this approach, we first prove that TCP performs competitively against any all powerful adversary if it is given a constant times more bandwidth and either (a) some extra time, or (b) we assume that no job ....

....can be allocated a variable (real) number of processors. Within these abstractions, the problem of scheduling bandwidth to a number of transmission sessions is identical to that of scheduling a number of processors to a set of parallelizable jobs. The latter problem has a rich history of results [MPT94, KP00, Edm99, EP01]. This paper applies and extends those results to the former problem. 3.1 The scheduling problem We assume that there is a single bottleneck in our network which causes all data losses. In reality this bottleneck may be a link or a router. We assume that the bottleneck has a maximum capacity ....

[Article contains additional citation context not shown here]

Bala Kalyanasundaram and Kirk Pruhs. Speed is as powerful as Clairvoyance. Journal of the ACM, 47(4):617--643, 2000.

....algorithm. Notice that when k is large, such performance guarantee is not satisfactory. In recent years, a plausible approach to obtaining better performance guarantee without making assumption on future inputs is to allow the online scheduler to have more resources than the adversary (e.g. [4, 6, 8, 10, 14, 16]) Speci cally, we would like to compare the online scheduler using a faster processor or more than one (unit speed) processors against an adversary using a unit speed processor. Intuitively, the additional resources are needed to compensate the online scheduler for the lack of future information. ....

....resources are needed to compensate the online scheduler for the lack of future information. The key question is whether a moderate amount of additional resources can provide satisfactory competitiveness. For the rm deadline scheduling problem, the pioneer work of Kalyanasundaram and Pruhs [10] showed that the competitive ratio can be improved to a constant independent of the importance ratio k when the online scheduler is given a moderately faster processor; precisely, they gave an algorithm called Slacker, which, if given a speed (1 2) processor for any 0, is (1 ....

[Article contains additional citation context not shown here]

Bala Kalyanasundaram and Kirk R. Pruhs. Speed is as powerful as clairvoyance. Journal of the ACM, 47(4):617643, 2000.

.... bound is tight as matching algorithms are also known [1, 9, 14] In recent years, there are a number of exciting results on improving performance guarantee without making assumption on future inputs; the basic idea is to allow the online scheduler to have more resources than the adversary (e.g. [3, 6 8, 10 12]) For the single processor deadline scheduling problem, the pioneer work of Kalyanasundaram and Pruhs [8] implies that the competitive ratio can be reduced arbitrarily if the on line scheduler is given a faster processor (e.g. the competitive ratio is roughly 2 if the processor speed increases ....

.... on improving performance guarantee without making assumption on future inputs; the basic idea is to allow the online scheduler to have more resources than the adversary (e.g. 3, 6 8, 10 12] For the single processor deadline scheduling problem, the pioneer work of Kalyanasundaram and Pruhs [8] implies that the competitive ratio can be reduced arbitrarily if the on line scheduler is given a faster processor (e.g. the competitive ratio is roughly 2 if the processor speed increases by 21 times) More recently, Lam and To [11] proved that optimality can be achieved with a moderate ....

Bala Kalyanasundaram and Kirk R. Pruhs. Speed is as powerful as clairvoyance. Journal of the ACM, 47(4):617--643, 2000.

....the total value of jobs that meet the deadlines; indeed, no algorithm can be (k) competitive, where k is the importance ratio of the jobs. Recent work, however, reveals that the competitive ratio can be improved to O(1) if the on line scheduler is equipped with a processor O(1) times faster [8]; furthermore, optimality can be achieved when using a processor O(log k) times faster [12] This paper presents a new on line algorithm for scheduling jobs with tight deadlines, which can achieve optimality when using a processor that is only O(1) times faster. 1 Introduction This paper is ....

.... ratio of ( where k is the importance ratio [9] In recent years, there are a number of exciting results on improving the performance guarantee without making assumption on future inputs; the basic idea is to allow the on line scheduler to have more resources than the adversary (e.g. [3, 6, 8, 13, 7, 11, 12]) For the single processor rm deadline scheduling problem, Kalyanasundaram and Pruhs [8] showed that the competitive ratio can be reduced signi cantly if the on line scheduler is given a faster processor. For instance, with a processor that is 32 times faster, the competitive ratio can be ....

[Article contains additional citation context not shown here]

Bala Kalyanasundaram and Kirk Pruhs. Speed is as powerful as clairvoyance. Journal of the ACM, 47(4):617-643, July 2000.

....the needs of all users in a fair way. This will be measured as the average time from when a job arrives until when it completes under the schedule, i.e. Avg i2J [c i Gamma a i ] This measure, which we denote by L(ALG s ) is standard both in systems community and in scheduling community [MPT94, KP00, Edm99, EP01]. It is typically referred to as the user perceived latency within the first community and as flow time within the second. TCP, the very widely used congestion control protocol, is in fact a very simple and natural scheduling algorithm. It is an online scheduler in that it does not know about ....

Bala Kalyanasundaram and Kirk Pruhs. Speed is as powerful as Clairvoyance. Journal of the ACM, 47(4):617--643, 2000. 15

....number of processors. Within these abstractions, these two problems are identical the problem of scheduling bandwidth to a number of transmission sessions is identical to that of scheduling a number of processors to a set of parallelizable jobs. The latter problem has a rich history of results [MPT94, KP00, Edm99, EP01]. This paper applies and extends those results to the former problem. 3.1 The scheduling model We assume that there is a single bottleneck (or point of congestion) in our network which causes all data losses. In reality this bottleneck may be the capacity of a link or a router. We assume that ....

....Since the bottleneck has capacity sB, i2J t b i;t sB. A job of length l i completes at time c i if the algorithm allocates enough bandwidth so that R t2[a i ;c ] b i;t = l i . We use the flow time L(ALG s ) of a scheduling algorithm ALG s as a measure of its fairness. The flow time [MPT94, KP00, Edm99, EP01] is the average time between the arrival time and completion time of a job, i.e. Avg i2J [c i Gammaa i ] This measure is sometimes called the user perceived latency in the Systems community. As mentioned before, we measure the performance of an algorithm by its competitive ratio which is ....

[Article contains additional citation context not shown here]

Bala Kalyanasundaram and Kirk Pruhs. Speed is as powerful as Clairvoyance. Journal of the ACM, 47(4):617--643, 2000.

....for the case when one of the player is subject to additional constraints. The extension is applicable to several contexts in the area of randomized algorithms, including multi objective optimization problems [20] performance tail of randomized algorithms [14] the resource augmentation method [11], smoothed analysis [23] and loose competitiveness [27] The rest of this section discusses the consequences of our main duality result for algorithm design and analysis. Algorithms. Several optimization problem naturally lend themselves to a multi objective formulation, where algorithms can ....

....The ability to increase the adversary cost immediately leads us to compare upper bounded adversary with the resource augmentation method (RAM) and with smoothed analysis. RAM and smoothed analysis. In the RAM, the algorithm resources are augmented as compared to those given to the adversary [11]. Thus, the resource augmentation causes the adversary cost to increase by d(x) on input x. On the other hand, if we were allowed to increase arbitrarily the algorithm resources, we would often obtain the trivial result that the algorithm is super optimal. Therefore, it is critical that the amount ....

[Article contains additional citation context not shown here]

Bala Kalyanasundaram and Kirk Pruhs. Speed is as powerful as clairvoyance. J. ACM, 47(4):617-643, 2000.

No context found.

Bala Kalyanasundaram and Kirk Pruhs, "Speed is as powerful as clairvoyance," Journal of the ACM, vol. 47, no. 4, pp. 617--643, 2000.

No context found.

Bala Kalyanasundaram and Kirk Pruhs. Speed is as powerful as clairvoyance. Journal of the ACM, 47(4):214-221, 2000.

No context found.

Bala Kalyanasundaram and Kirk Pruhs. Speed is as powerful as clairvoyance. Journal of the ACM, 47(4):214-221, 2000.

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