M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Journal of Scheduling, 3(5):273-288, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and or a fee. STOC 02, May 19 21, 2002, Montreal, Quebec, Canada. Copyright 2002 ACM 1 58113 495 9 02 0005 . 5.00. petitive analysis [6] and extra resource analysis [2, 5, 7, 13, 16, 17, 18, 20], to study several interesting problems related to the design of on line schedulers on di erent VOD systems and give provably e ective solutions for these problems. In a VOD system, a hot video is often requested over a short period of time (say, Friday 7 p.m. to 9 p.m. Due to the large number ....

....video comprises 120 time units. The total bandwidth for serving both clients is reduced from 240 to 121. ness, and Chan et al. 9] improved their result to obtain a 5 competitive scheduler. The second question is closely related to the extra resource analysis of on line algorithms (see, e.g. [2, 5, 7, 13, 16, 17, 18, 20]) As mentioned before, we have an on line scheduler which is 3 competitive for 20 skimming. Our result suggests that by taking advantage of extra resources, there is an online scheduler for 5 skimming which is 1.2375 competitive relative to 20 skimming. Furthermore, suppose we want to determine ....

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. J. Scheduling, 3(5):273-288, 2000.

.... cannot match or be competitive against the o ine adversary [2] In recent years, a plausible approach to achieving better performance guarantee for online scheduling (without restricting the inputs) is to allow the online scheduler to use a faster processor than the o ine adversary (e.g. [3, 5, 6, 9, 12,16, 18]) Intuitively, we use a faster processor to compensate the online scheduler for the lack of future information. The key question is whether a A preliminary version of this paper appeared in the Proceedings of the 14th ACM Annual Symposium on Parallel Algorithms and Architectures, 2002. ....

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Journal of Scheduling, 3(5):273-288, 2000.

.... the performance cannot match or be competitive against the o#ine adversary [2] In recent years, a plausible approach to studying better performance guarantee for online scheduling (without restricting the inputs) is to allow the online scheduler to use a faster processor than the o#ine adversary [3,5,7,10,14,16]. Intuitively, we need a faster processor to compensate the online scheduler for the lack of future information. The key question is whether a moderate amount of extra speed can # Department of Computer Science, University of Maryland, College Park, MD 20742, USA (cykoo cs.umd.edu) Department ....

....passed to but not completed by Band 2 (i.e. I A1 A2 ) Furthermore, we need the following definition. Definition 2. The span of a job J is the period [r(J) d(J) and the span of a set of jobs is the union of the spans of all the jobs in S. e.g. the union of the spans [3, 6] and [5, 8] is [3, 8] Furthermore, let sp(S) be the total time included in the span of S. Theorem 9. p(A2 ) sp(I # ) Before proving Theorem 9, we note that Theorem 9 guarantees that EDF MSp is a four processor optimal algorithm for scheduling jobs with value densities in the range [1, 2] Corollary ....

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. In Proc. 11th ACM-SIAM SODA, pages 560--561, 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. ....

Mark Brehob, Eric Torng, and Patchrawat Uthaisombut. Applying extra-resource analysis to load balancing. In Proceedings of the Eleventh Annual ACM-SIAM Symposium on Discrete Algorithms, pages 560561, San Francisco, California, 911 January 2000.

....o line algorithm. Indeed, in most settings, no online algorithm has this sort of performance guarantee [2, 8] In recent years, a plausible approach to studying performance guarantee for online scheduling without restricting the inputs is to allow the online scheduler to use faster processors [1, 3, 5, 9, 10, 13, 14]. Intuitively, we want to study how e ective faster processors can compensate the online scheduler for the lack of future information. Phillips et al. 14] were able to extend the optimality of edf to the underloaded, multiprocessor setting by allowing the online scheduler to use double speed ....

Mark Brehob, Eric Torng, and Patchrawat Uthaisombut. Applying extra-resource analysis to load balancing. In Proceedings of the Eleventh Annual ACM-SIAM Symposium on Discrete Algorithms, pages 560561, San Francisco, California, 911 January 2000.

.... the performance cannot match or be competitive against the o ine adversary [2] In recent years, a plausible approach to studying better performance guarantee for online scheduling (without restricting the inputs) is to allow the online scheduler to use a faster processor than the o ine adversary [3, 5, 8, 11, 15, 17]. Intuitively, we use a faster processor to compensate the online scheduler for the lack of future information. The key question is whether a moderate amount of extra speed can lead to satisfactory competitiveness. Kalyanasundaram and Pruhs [11] are the rst to exploit a faster processor to derive ....

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. In Proceedings of the Eleventh Annual ACM-SIAM Symposium on Discrete Algorithms, pages 560-561, 2000.

....time was considered in numerous papers [10, 11, 4, 14, 1, 9, 8] Other papers also considered di erent functions of the completion times [2] but never the starting times. Resource augmentation for scheduling of jobs one by one was also considered with the maximum completion time goal function [6, 3]. However, to the best of our knowledge, no previous work on the above goal function exists. We show the following results for the competitive ratio on identical machines: The greedy algorithm, which assigns each job to the least loaded machine, has competitive ratio (log #) The greedy ....

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Journal of Scheduling, 3(5):273-288, 2000.

.... c m;n (A) Specifically, c m;n (A) max A n ( OPTm ( The classical case of n = m was considered in a series of papers [11, 12, 3, 15, 1] The best upper bound is 1:923 due to Albers [1] and the best lower bound is 1:853 [10] based on [1] The case n m was introduced by Brehob et al. [5]. They showed that no matter how many machines the on line algorithm has, it can never perform optimally: c m;n (A) 1 for all n m 2. However, one would expect that for any reasonable algorithm A, c m;n (A) will approach 1 when t = n=m tends to in nity. In fact, 5] showed that the greedy ....

....introduced by Brehob et al. [5] They showed that no matter how many machines the on line algorithm has, it can never perform optimally: c m;n (A) 1 for all n m 2. However, one would expect that for any reasonable algorithm A, c m;n (A) will approach 1 when t = n=m tends to in nity. In fact, [5] showed that the greedy algorithm has a competitive ratio which approaches 1 in a rate depending linearly on 1=t. In contrast, we design an algorithm with a competitive ratio which approaches 1 in a rate depending exponentially on t. More speci cally, we give 2 PERMANENT TASKS 3 an algorithm of ....

[Article contains additional citation context not shown here]

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Manuscript, 1999.

....c m,n (A) Specifically, c m,n (A) max # A n (#) OPTm (#) The classical case of n = m was considered in a series of papers [11, 12, 3, 15, 1] The best upper bound is 1.923 due to Albers [1] and the best lower bound is 1. 853 [10] based on [1] The case n m was introduced by Brehob et al. [5]. They showed that no matter how many machines the on line algorithm has, it can never perform optimally: c m,n (A) 1 for all n m # 2. However, one may expect that for reasonable algorithms c m,n (A) would approach 1 when t = n m increases. In fact, 5] showed that the greedy algorithm has a ....

....n m was introduced by Brehob et al. [5] They showed that no matter how many machines the on line algorithm has, it can never perform optimally: c m,n (A) 1 for all n m # 2. However, one may expect that for reasonable algorithms c m,n (A) would approach 1 when t = n m increases. In fact, [5] showed that the greedy algorithm has a competitive ratio which approaches 1 in a rate depending linearly on 1 t. In contrast, while the greedy algorithm has a competitive ratio which approaches 1 in a rate depending linearly on 1 t, we design a non greedy algorithm whose competitive ratio ....

[Article contains additional citation context not shown here]

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Manuscript, 1999.

.... (A) Specifically, c m;n (A) max oe A n (oe) OPTm (oe) The classical case of n = m was considered in a series of papers [11, 12, 3, 15, 1] The best upper bound is 1:923 due to Albers [1] and the best lower bound is 1:853 [10] based on [1] The case n m was introduced by Brehob et al. [5]. They showed that no matter how many machines the on line algorithm has, it can never perform optimally: c m;n (A) 1 for all n m 2. However, one may expect that for reasonable algorithms c m;n (A) would approach 1 when t = n=m increases. In fact, 5] showed that the greedy algorithm has a ....

....case n m was introduced by Brehob et al. [5] They showed that no matter how many machines the on line algorithm has, it can never perform optimally: c m;n (A) 1 for all n m 2. However, one may expect that for reasonable algorithms c m;n (A) would approach 1 when t = n=m increases. In fact, [5] showed that the greedy algorithm has a competitive ratio which approaches 1 in a rate depending linearly on 1=t. In contrast, while the greedy algorithm has a competitive ratio which approaches 1 in a rate depending linearly on 1=t, we design a non greedy algorithm whose competitive ratio ....

[Article contains additional citation context not shown here]

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Manuscript, 1999.

No context found.

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Journal of Scheduling, 3(5):273-288, 2000.

No context found.

Mark Brehob, Eric Torng, and Patchrawat Uthaisombut. Applying extra-resource analysis to load balancing. In Proceedings of the Eleventh Annual ACM-SIAM Symposium on Discrete Algorithms, pages 560--561, San Francisco, California, January 2000.

No context found.

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Journal of Scheduling, 3(5):273--288, 2000. 8

No context found.

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Journal of Scheduling, 3(5):273--288, 2000.

No context found.

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. Journal of Scheduling, 3(5):273--288, 2000.

No context found.

M. Brehob, E. Torng, and P. Uthaisombut. Applying extra-resource analysis to load balancing. In Proceedings of the Eleventh Annual ACM-SIAM Symposium on Discrete Algorithms, pages 560-561, 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