Download:
|
by Susanne Albers, Bianca Schroder
ACM Journal of Experimental Algorithmics
http://ls2-www.informatik.uni-dortmund.de/~albers/wae.ps.gz
Add To MetaCart
Abstract:
We present the first comprehensive experimental study of online algorithms for Graham's scheduling problem. Graham's scheduling problem is a fundamental problem in scheduling theory where a sequence of jobs has to be scheduled on m identical parallel machines so as to minimize the makespan. Graham gave an elegant algorithm that is (2 1/m)-competitive. Recently a number of new online algorithms were developed that achieve competitive ratios around 1.9. Since competitive analysis can only capture the worst case behavior of an algorithm a question often asked is: Are these new algorithms geared only towards a pathological case or do they perform better in practice, too? We address this question by analyzing the algorithms on various job sequences. We have implemented a general testing environment that allows a user to generate jobs, execute the algorithms on arbitrary job sequences and obtain a graphical representation of the results. In our actual tests, we analyzed the algorithms (1) on real world jobs and (2) on jobs generated by probability distributions. It turns out that the performance of the algorithms depends heavily on the characteristics of the respective work load. On job sequences that are generated by standard probability distributions, Graham's strategy is clearly the best. However, on the real world jobs the new algorithms often outperform Graham's strategy. Our experimental study confirms theoretical results and gives some new insights into the problem. In particular, it shows that the techniques used by the new online algorithms are also interesting from a practical point of view. 1
Citations
|
666
|
The Art of Computer Systems Performance Analysis
– Jain
- 1991
|
|
239
|
Bounds for certain multiprocessing anomalies
– Graham
- 1966
|
|
165
|
Scheduling to Minimize Average Completion Time: Off-line and On-line Approximation Algorithms
– Hall, Schulz, et al.
- 1997
|
|
122
|
Scheduling parallel machines online
– Shmoys, Wein, et al.
- 1995
|
|
102
|
New algorithms for an ancient scheduling problem
– Bartal, Fiat, et al.
- 1995
|
|
81
|
Amortized eciency of list update and paging rules
– Sleator, E
- 1985
|
|
73
|
Better bounds for online scheduling
– Albers
- 1999
|
|
64
|
A better algorithm for an ancient scheduling problem
– Karger, Phillips, et al.
- 1996
|
|
43
|
An on-line scheduling heuristic with better worst case ratio than Graham’s list scheduling
– Galambos, Woeginger
- 1993
|
|
34
|
A Lower Bound for Randomized On-Line Multiprocessor Scheduling
– Sgall
- 1997
|
|
32
|
An Experimental Study of LP-Based Approximation Algorithms for Scheduling Problems
– SAVELSBERGH, UMA, et al.
- 1998
|
|
28
|
A better lower bound for on-line scheduling
– Bartal, Karloff, et al.
- 1994
|
|
21
|
The effect of heavy-tailed job size distributions on computer system design
– Harchol-Balter
- 1999
|
|
21
|
On-Line Scheduling. In Online Algorithms: The State of the
– Sgall
- 1998
|
|
12
|
On the performance of on-line algorithms for particular problems
– Faigle, Kern, et al.
- 1989
|
|
10
|
Lower bounds for randomized online scheduling
– Chen, Vliet, et al.
- 1994
|
|
7
|
Minimizing expected makespans on uniform processor systems
– Coffman, Flatto, et al.
- 1987
|
|
7
|
editors. Job Scheduling Strategies for Parallel
– Feitelson, Rudolph
- 1995
|
|
5
|
A note on expected makespans for largest-first sequences of independent tasks on two processors
– Jr, Frederickson, et al.
- 1984
|
|
4
|
Expected makespans for largest-first multiprocessor scheduling
– Jr, Flatto, et al.
- 1984
|
