Anna R. Karlin, M.S. Manasse, L.A. McGeoch, and S. Owicki. Competitive Randomized Algorithms For Nonuniform Problems. Algorithmica, 11(6):542--71, June 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is defined to be the average of its costs over all of the possible series of random choices. Competitiveness is defined as before, but it uses this modified definition of cost. For a number of different problems, it has been shown that the competitive factor can be reduced by the use of randomness [17, 27]. We have presented and analyzed new strongly competitive algorithms for replication and migration problems that arise in the management of distributed shared memory for multiprocessor systems. These algorithms are applicable to many existing and proposed multiprocessor architectures. The proofs ....

Anna Karlin, Mark Manasse, Lyle McGeoch, and Susan Owicki. Competitive Randomized Algorithms for Non-Uniform Problems. In Proceedings, Symposium on Discrete Algorithms, 1990.

....policy, primarily due to the exponential interarrival request assumption. This policy both misses some long idle periods, and tends to shut down too agressively, as can be seen from its very short average sleep time, T as . Our policy also outperforms one based on the competitive algorithm (CA) [1] that is guaranted to yield a policy that consumes at worst twice the minimum amount of power consumed by the policy computed with perfect knowledge of the user behavior. Table 1. Measurement Comparison Algorithm Pwr (W) Tas (s) oracle policy 0.33 118 TISMDP 0.40 81 CA (5.43s timeout) 0.44 ....

L. M. A. Karlin, M. Manesse and S. Owicki. Competitive randomized algorithms for nonuniform problems. Algorithmica, pages 542--571, 1994.

....these heuristics can be analyzed and propose algorithms that are independent of the input distribution. This framework is based on the notion of competitive analysis. Competitive analysis has been used as a technique to analyze various on line problems [23] 22] 21] 20] 18] 19] In [8] [7], the authors analyze the spin block problem which is similar to the power management problem without any latency consideration. The algorithms and proofs presented in this paper have been adapted from those works. There are signi cant changes in the problem formuation with latency consideration ....

....spin block problem which is similar to the power management problem without any latency consideration. The algorithms and proofs presented in this paper have been adapted from those works. There are signi cant changes in the problem formuation with latency consideration which the authors in [8] [7] have not addressed. On the whole, this paper makes ve major contributions. The rst contribution is the introduction of a formal analysis technique called competitive analysis to the power management problem. Competitive analysis as applied to power management does not depend on trace patterns ....

[Article contains additional citation context not shown here]

Karlin A. R., Manasse M.S., McGeoch L.A., Owicki S. Competitive Randomized Algorithms for Nonuniform Problems. Algorithmica, vol. 11, no 6, pp 542-571, June 1994

....maximizes performance under power constraint) is a constrained optimisation problem. The most common power management policy at the system level is a timeout policy implemented in most operating systems. The drawback of this policy is that it wastes power while waiting for the timeout to expire [8,9]. Predictive policies developed for interactive terminals [10,11] force the transition to a low power state as soon as a component becomes idle if the predictor estimates that the idle period will last long enough. An incorrect estimate can cause both performance and energy penalties. Both timeout ....

....keeping the power consumption down. From our experience with the user interaction with the hard disk, our algorithm performs well, thus giving us low power consumption with still good performance. In comparison, Karlin s policy consumes 10 more power and has worse performance. Karlin s algorithm [8] guarantees to yield a policy that consumes at worst twice the minimum amount of power consumed by the policy computed with perfect knowledge of the user behaviour. In addition, our policy consumes 1.7 times less power than the default Windows timeout policy of 120s and 1.4 times less power than ....

[Article contains additional citation context not shown here]

A. Karlin, M. Manesse, L. McGeoch and S. Owicki, "Competitive Randomized Algorithms for Nonuniform Problems", Algorithmica, pp. 542--571, 1994.

.... Several excellent reviews of power minimiza tion techniques are presented by Pedram [6] Devadas and Malik [7] and Najm [8] The authors would like to acknowledge support from National Science Foundation award numbers MIP 95 01615 (CAREER) and CCR 9806898, and from DARPA grant DABT63 98 C 0045. tParts of this work was done while this author was working for CynApps Inc. Santa Clara, CA 94043. Low power VLSI design can be achieved at various levels of abstractions during the design process. These include the system level, behavioral level, the RTL level, and the gate level. Most the ....

....and Malik [7] and Najm [8] The authors would like to acknowledge support from National Science Foundation award numbers MIP 95 01615 (CAREER) and CCR 9806898, and from DARPA grant DABT63 98 C 0045. tParts of this work was done while this author was working for CynApps Inc. Santa Clara, CA 94043. Low power VLSI design can be achieved at various levels of abstractions during the design process. These include the system level, behavioral level, the RTL level, and the gate level. Most the techniques in the literature are focused at the RTL level. This paper focuses on the system level ....

[Article contains additional citation context not shown here]

Karlin A. R., Manasse M.S., McGeoch L.A., Owicki S. Competitive Randomized Algorithms for Nonuniform Problems. Algorithmtea, vol. 11, no 6, pp 542-571, June 1994

.... If disk interarrival times are independently drawn from some unknown stationary distribution, as the first argument states, then no matter what this distribution, there exists an inactivity threshold that incurs a cost no more than e (e 1) times that of the optimal off line transition strategy [23]. One could find this threshold by keeping track of all interarrival times so that the distribution, and thus the ideal threshold, could be deduced. One algorithm of that type, using constant space, builds up a picture of the past interarrival time distribution in the following indirect way [24] ....

....can do well no matter when disk accesses occur. One such strategy chooses a new random threshold after every disk access according to the cumulative distribution function where c is the number of seconds it takes the running motor to consume the same amount of energy it takes to spin up the disk [23]. This strategy has been proven ideal among strategies having no knowledge of the arrival process. Note, however, that almost all transition strategies described in this article do purport to know something about the arrival process, and thus are capable of beating this strategy. In other words, ....

A. R. Karlin, M. S. Manasse, L. A. McGeoch, and S. Owicki, "Competitive randomized algorithms for nonuniform problems," Algorithmica, vol. 11, no. 6, June 1994, pp. 542--571.

....scheduling algorithm uses all three steps. After the execution orders are determined, a 2 competitive power manager (2CPM) decides power states. A 2CPM is an on line power management algorithm using as the timeout value; it consumes at most twice of power compared to an oracle power manager [7]. Table 5 summarizes the simulation results. These devices consume totally 30 W in their working states. Approximately 10 power can be saved when applying power management to the base scheduling. Compared to the base scheduling, additional 20 and 33 power can be saved by the grouping and the ....

A. Karlin, M. Manasse, L. McGeoch, and S. Owicki. Competitive Randomized Algorithms for Nonuniform Problems. Algorithmica, 11(6):542--571, June 1994.

....locks. These allow the expression and efficient execution of finegrain parallelism and lead to more frequent synchronization operations and shorter wait times compared to programs on small scale machines that use less scalable synchronization methods. In previous analytical work, Karlin et al. [19] present an efficient randomized waiting algorithm that achieves an optimal on line competitive factor of e (e 1) against a wea adversary. See Section 4 for a definition of competitive factors, adversaries and optimality. We prove in this paper that if we restrict the adversary by fixing the ....

....choose the correct waiting mechanism at all times. An on line algorithm is strogly competitive, and thus optimal, if it possesses the smallest possible competitive factor. The cost of a waiting algorithm depends on the sequence of wait times presented to it by an adversary. Using terminology in [19], a strotg adversary is one that chooses wait times in response to previous choices of Lvoll by the waiting algorithm. A weak adversary is one that chooses wait times without considering previous choices of Lvoli by the waiting algorithm. We can easily achieve a 2 competitive waiting algorithm ....

[Article contains additional citation context not shown here]

A. Karlin, M. Manasse, L. McGeoch, and S. Owicki. Competitive Randomized Algorithms for Non-Uniform Problems. In Proceedings 1st Annual ACM-SIAM Symposium on Discrete Algorithms, pages 301 309, January 1990.

....Fig. 3, a timeout policy shuts down the device at . These policies assume that if a device is idle longer than then it will remain idle for at least [7] When equals to the break even time, the device consumes at most twice power compared to a perfect policy; this is called a 2 competitive policy [14]. An obvious drawback of timeout policies is the wasted energy during the timeout period if is large. Predictive policies explicitly predict the length of an idle period before it starts. If the predicted length is larger than , the device is shut down immediately after it becomes idle. These ....

....another requester is created two min later. Each workload runs for two h. C. Policy Comparison Seven policies are compared. The last two policies are used for the third workload in which requests are generated with timers: 1) no power management; 2) timeout of three minutes; 3) 2 competitive [14]; 4) exponential average (16) 5) process based; 6) predictive wakeup; 7) scheduling for flexible timers with predictive wakeup. We compare these policies by five criteria, including power and performance. Power is determined by the time in the working state and the number of state ....

A. R. Karlin, M. S. Manasse, L. A. McGeoch, and S. Owicki, "Competitive randomized algorithms for nonuniform problems," Algorithmica, vol. 11, no. 6, pp. 542--571, June 1994.

.... a certain time and then blocks if synchronization is not yet achieved [9] When waiting for a lock, spinning for a time equal to CS is 2competitive, meaning that it results in an execution that is at most a factor of two from that of an optimal execution in which wait times are known in advance [6]. This is the best possible result for a deterministic algorithm, but a randomized algorithm can achieve a competitive factor of e e 1 1:58. The advantage of two phase blocking has also been demonstrated experimentally [7] Different synchronization mechanisms, however, have different ....

A. Karlin, M. S. Manasse, L. A. McGeoch, and S. Owicki. Competitive randomized algorithms for non-uniform problems. In Proc. 1st ann. ACM-SIAM symp. Discrete Algorithms, pages 301309, January 1990.

.... blocks if synchronization is not yet achieved [10] When waiting for a lock, spinning for a time equal to the context switch overhead is 2 competitive, meaning that it results in an execution that is at most a factor of two from that of an optimal execution in which wait times are known in advance [6]. This is the best possible result for a deterministic algorithm, but a randomized algorithm can achieve a competitive factor of e e 1 1:58. The advantage of two phase blocking has also been demonstrated experimentally [7] Di erent synchronization mechanisms, however, have di erent ....

A. Karlin, M. S. Manasse, L. A. McGeoch, and S. Owicki, \Competitive randomized algorithms for non-uniform problems". In Proc. 1st ann. ACM-SIAM symp. Discrete Algorithms, pp. 301-309, January 1990.

.... and the requests alternate between A and B) For the page replication problem, there are optimal 2 competitive deterministic [4] and (1 1 N ) N (1 1 N ) N Gamma1 competitive randomized algorithms against an oblivious adversary [1, 8] A similar bound was obtained by Karlin et al. [6] for the problem of two servers on a (1; N; N) triangle. In this paper, we consider the Bahncard Problem which contains the SRP as a special case (another generalization of the SRP was given in [2] The Bahncard is a railway pass of the Deutsche Bundesbahn (the German railway company) 2 . It ....

A.R. Karlin, M.S. Manasse, L.A. McGeoch, and S. Owicki. Competitive randomized algorithms for non-uniform problems. In Proceedings of the 1st ACMSIAM Symposium on Discrete Algorithms (SODA'90), pages 301--309, 1990.

....c(x) 0 for x k and c(x) 1 for x k, then we are back at classical paging with cache size k. Moreover, our problem captures some features of classical renting versus buying problems (the most simple instance of which is the well known ski rental problem (see Karlin, Manasse, McGeoch Owicki [6]) Imreh Noga [4] analyze machine scheduling problems where the online algorithm may adjust its resources at an additional cost. Notation. For a xed algorithm A and a given request sequence , we denote by fault A ( the number of page faults that A incurs on , and we denote by cache A ( ....

A. Karlin, M. Manasse, L. McGeoch, and S. Owicki. Competitive randomized algorithms for nonuniform problems. Algorithmica 11, 542-571, 1994.

....the two objective functions that we consider and the circumstances under which they are appropriate measures of performance. In Section 6. 3, we describe the key differences between the on line problem considered here and some classic on line problems such as the rent to buy (or spin block) problem [57, 50] and the snoopy caching problem [50] Finally, in Section 6.4, we summarize the results contained in this part of the thesis. 6.1 Motivation Most TCP implementations used today employ some sort of acknowledgment delay mechanism. Delaying acknowledgments allows the TCP to acknowledge multiple ....

....and the circumstances under which they are appropriate measures of performance. In Section 6. 3, we describe the key differences between the on line problem considered here and some classic on line problems such as the rent to buy (or spin block) problem [57, 50] and the snoopy caching problem [50]. Finally, in Section 6.4, we summarize the results contained in this part of the thesis. 6.1 Motivation Most TCP implementations used today employ some sort of acknowledgment delay mechanism. Delaying acknowledgments allows the TCP to acknowledge multiple incoming data segments with a single ....

[Article contains additional citation context not shown here]

A. R. Karlin, M. S. Manasse, L. A. McGeoch, and S. Owicki. Competitive randomized algorithms for non-uniform problems. In Proceedings of the First Annual ACM-SIAM Symposium on Discrete Algorithms, pages 301--309, 1990.

No context found.

Anna R. Karlin, Mark S. Manasse, Lyle A. McGeoch, and Susan Owicki. \Competitive Randomized Algorithms for Non-Uniform Problems." In Proceedings of the First Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 301-309, San Francisco, California, 22-24 January 1990.

....redistribute to lists, requires prior specific permission and or a fee. STOC 01 July 6 8, 2001, Hersonissos, Crete, Greece. Copyright 2001 ACM 1 58113 349 9 01 0007 . 5.00. of e= e Gamma 1) which is about 1. 58) in the limit as the ratio between the buy cost and the rent cost becomes large [9]. Dynamic TCP acknowledgment A stream of packets arrive at a destination. The TCP protocol requires that these packets be acknowledged. However, the possibility exists of using a single acknowledgement packet to simultaneously acknowledge multiple outstanding packets, thereby reducing the ....

Anna R. Karlin, Mark S. Manasse, Lyle A. McGeoch, and Susan Owicki. "Competitive Randomized Algorithms for Non-Uniform Problems". In Proceedings of the First Annual ACM-SIAM Symposium on Discrete Algorithms, pages 301-309, San Francisco, California, 22-24 January

....marking algorithm, but achieves the optimal competitive factor of H k . For deterministic server problems all evidence indicates that the optimal competitive factor is k, and is therefore independent of the distances in the graph [4, 5, 11] This is not true in the randomized case. Karlin et al. [8] have shown that for two servers in a graph that is an isosceles triangle the best competitive factor that can be achieved is a constant that approaches e= e Gamma 1) 1:582 as the length of the similar sides go to infinity. This contrasts with the uniform 3 vertex, 2 server problemfor which ....

A. R. Karlin, M. S. Manasse, L. A. McGeoch, and S. Owicki. Competitive randomized algorithms for non-uniform problems. In First Annual ACM-SIAM Symposium on Discrete Algorithms, pages 301--309, San Francisco, January 1990.

No context found.

Anna R. Karlin, M.S. Manasse, L.A. McGeoch, and S. Owicki. Competitive Randomized Algorithms For Nonuniform Problems. Algorithmica, 11(6):542--71, June 1994.

No context found.

Karlin, A. R., Manasse, M. S., McGeoch, L. A., and Owicki, S. Competitive randomized algorithms for nonuniform problems. Algorithmica, 11(6):542--571, June 1994.

No context found.

A. Karlin, M. Manesse, L. McGeoch and S. Owicki, "Competitive Randomized Algorithms for Nonuniform Problems", Algorithmica, pp. 542--571, 1994.

No context found.

A. Karlin, M. Manasse, L. McGeoch, and S. Owicki. Competitive Randomized Algorithms for Nonuniform Problems. Algorithmica, 11(6):542--571, June 1994.

No context found.

A. Karlin, M. Manasse, L. McGeoch, and S. Owickim, "Competitive Randomized Algorithms for Non-Uniform Problems," Algorithmica, vol. 11, no. 6, pp. 542-571, June 1994.

No context found.

A. Karlin, M. Manesse, L. McGeoch, and S. Owicki, "Competitive randomized algorithms for nonuniform problems," Algorithmica, vol. 11, no. 6, pp. 542--571, June 1994.

No context found.

A. Karlin, M. Manasse, L. McGeoch, and S. Owicki, "Competitive randomized algorithms for nonuniform problems," Algorithmica, vol. 11, no. 6, pp. 542--571, June 1994.

No context found.

A. Karlin et al., "Competitive Randomized Algorithms for Non-Uniform Problems," Algorithmica, vol. 11, no. 6, June 1994, pp. 542-571.

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