## Work-Time-Optimal Parallel Algorithms for String Problems (Extended Abstract) (1995)

Venue: | In Proc. 27th ACM Symp. on the Theory of Computing |

Citations: | 5 - 2 self |

### BibTeX

@INPROCEEDINGS{Czumaj95work-time-optimalparallel,

author = {Arthur Czumaj and Zvi Galil and Wojciech Plandowski},

title = {Work-Time-Optimal Parallel Algorithms for String Problems (Extended Abstract)},

booktitle = {In Proc. 27th ACM Symp. on the Theory of Computing},

year = {1995},

pages = {713--722}

}

### OpenURL

### Abstract

) Artur Czumaj Zvi Galil y Leszek G¸asieniec z Kunsoo Park x Wojciech Plandowski -- Abstract A parallel algorithm is work-optimal if it uses the smallest possible work; a work-optimal algorithm is worktime -optimal if it also uses the smallest possible time. We design work-time-optimal algorithm for a number of string processing problems on the EREW-PRAM and the hypercube. They include string matching and two dimensional pattern matching. No such algorithms have been known before for any of these problems. 1 Introduction We call a parallel algorithm work-optimal if it has smallest possible work. Notice that this definition is stricter than the one requiring only the same work as the best known sequential algorithm and it requires proving a lower bound. In most cases work-optimality means either linear work or O(n log n) work because no higher lower bounds are known. We call a work-optimal algo- Heinz Nixdorf Institute, University of Paderborn, D-33095 Paderborn, Germany....

### Citations

1332 |
Introduction to parallel algorithms and architectures: arrays, trees, hypercubes
- Leighton
- 1992
(Show Context)
Citation Context ...e hypercube family which includes in addition the butterfly, the Benes, the shuffle-exchange and the de Bruijn networks as well as the Cube-Connected-Cycles, all of which are constant degree networks =-=[22]-=-. In this paper we consider the following eight string processing problems. 1. text search of string matching given a preprocessed pattern 2. computing witnesses for a fraction of the pattern 3. text ... |

285 | Parallel merge sort
- Cole
- 1988
(Show Context)
Citation Context ...ound for computing the AND function and for broadcasting on the EREWPRAM does not leave much space for such algorithms. In fact, on the EREW-PRAM we are aware only of wtoptimal algorithms for sorting =-=[11]-=- and some algorithms on planar graphs [19]. On the hypercube the situation is even worse. Since even for such a basic problem as routing no wt-optimal algorithm is known, any problem with a wt-optimal... |

95 |
Upper and lower time bounds for parallel random access machines without simultaneous writes
- Cook, Dwork, et al.
- 1986
(Show Context)
Citation Context ...n) lower bound for the work of Problem 7 follows from [23]. A logarithmic lower bound for all these problems on the CREW-PRAM follows immediately from the logarithmic lower bound for the AND function =-=[9]-=-. The only two cases where wt-optimal algorithms are known on the CREW-PRAM are for Problem 1 [27] and Problem 3 [2]. Surprisingly, very few (deterministic) wt-optimal algorithms are known on the more... |

76 | Finding the maximum, merging, and sorting in a parallel computation model - SHILOACH, VISHKIN - 1981 |

72 |
An O(n log n) algorithm for finding all repetitions in a string
- Main, Lorentz
- 1984
(Show Context)
Citation Context ...n log n) work and O(log log n) time for Problem 7 [3]. The\Omega\Gammae/4 log n) lower bounds for the time follow from [7]. The\Omega\Gamma n log n) lower bound for the work of Problem 7 follows from =-=[23]-=-. A logarithmic lower bound for all these problems on the CREW-PRAM follows immediately from the logarithmic lower bound for the AND function [9]. The only two cases where wt-optimal algorithms are kn... |

71 |
Parallel Permutation and Sorting Algorithms and a New Generalized Connection Network
- Nassimi, Sahni
- 1982
(Show Context)
Citation Context ..., we can use sparse sorting to obtain a wt-optimal algorithm for Problem 1 on the EREW-PRAM. Sparse sorting is useful for implementations on the hypercube because it takes O(log n) time on the n-cube =-=[24, 25]-=-. The preprocessing of string matching and two dimensional pattern matching includes in addition to the witness computation (Problems 2 and 4) also the computation of deterministic samples. But this s... |

46 | Relations between concurrent-write models of parallel computation - Fich, Ragde, et al. - 1988 |

39 |
Two-dimensional periodicity and its application
- Amir, Benson
- 1992
(Show Context)
Citation Context ...an (m= log log m)-block. 5 New Two Dimensional Techniques Next we describe three new techniques that we need for handling Problems 3 and 4. We first recall the definition of two dimensional witnesses =-=[1]-=-. Let v be a two dimensional vector and place one copy of a two dimensional pattern P above another with displacement v of the origin. If the aligned parts match, then v is a (two dimensional) period.... |

38 | Truly Alphabet-Independent Two-Dimensional Pattern Matching - Galil, Park - 1992 |

32 | Optimal parallel algorithms for periods, palindromes and squares
- Apostolico, Breslauer, et al.
- 1992
(Show Context)
Citation Context ...e eight problems there are wt-optimal algorithms known on the CRCW-PRAM : linear work and constant time for Problems 1 [15] and 3 [12]; linear work and O(log log n) time for Problems 2 [6], 4 [12], 5 =-=[4]-=-, 6 [13], 8 [4]; O(n log n) work and O(log log n) time for Problem 7 [3]. The\Omega\Gammae/4 log n) lower bounds for the time follow from [7]. The\Omega\Gamma n log n) lower bound for the work of Prob... |

27 | An optimal o(log log n) time parallel string matching algorithm
- Breslauer, Galil
- 1990
(Show Context)
Citation Context ...s. For all these eight problems there are wt-optimal algorithms known on the CRCW-PRAM : linear work and constant time for Problems 1 [15] and 3 [12]; linear work and O(log log n) time for Problems 2 =-=[6]-=-, 4 [12], 5 [4], 6 [13], 8 [4]; O(n log n) work and O(log log n) time for Problem 7 [3]. The\Omega\Gammae/4 log n) lower bounds for the time follow from [7]. The\Omega\Gamma n log n) lower bound for t... |

25 | A lower bound for parallel string matching
- Breslauer, Galil
- 1992
(Show Context)
Citation Context ...r work and O(log log n) time for Problems 2 [6], 4 [12], 5 [4], 6 [13], 8 [4]; O(n log n) work and O(log log n) time for Problem 7 [3]. The\Omega\Gammae/4 log n) lower bounds for the time follow from =-=[7]-=-. The\Omega\Gamma n log n) lower bound for the work of Problem 7 follows from [23]. A logarithmic lower bound for all these problems on the CREW-PRAM follows immediately from the logarithmic lower bou... |

24 |
New Parallel-Sorting Schemes
- Preparata
(Show Context)
Citation Context ..., we can use sparse sorting to obtain a wt-optimal algorithm for Problem 1 on the EREW-PRAM. Sparse sorting is useful for implementations on the hypercube because it takes O(log n) time on the n-cube =-=[24, 25]-=-. The preprocessing of string matching and two dimensional pattern matching includes in addition to the witness computation (Problems 2 and 4) also the computation of deterministic samples. But this s... |

21 |
An Introduction to Parallel Algorithms (Addison-Wesley
- JaJa
- 1992
(Show Context)
Citation Context ...e CRCW-PRAM algorithms above to the ones for weaker models. For example, a T (n)-time CRCWPRAM algorithm on p processors can be simulated by an O(T (n) log p)-time EREW-PRAM algorithm on p processors =-=[20]-=-. (In our case log p = O(log n).) But unlesssT (n) is a constant we get time larger than O(log n) and in all cases we lose work-optimality. On the hypercube simulations give even worse results. This p... |

19 | Optimally fast parallel algorithms for preprocessing and pattern matching in one and two dimensions
- Cole, Crochemore, et al.
- 1993
(Show Context)
Citation Context ...In this abstract we only discuss the first four problems. For all these eight problems there are wt-optimal algorithms known on the CRCW-PRAM : linear work and constant time for Problems 1 [15] and 3 =-=[12]-=-; linear work and O(log log n) time for Problems 2 [6], 4 [12], 5 [4], 6 [13], 8 [4]; O(n log n) work and O(log log n) time for Problem 7 [3]. The\Omega\Gammae/4 log n) lower bounds for the time follo... |

19 |
A constant-time optimal parallel string-matching algorithm
- Galil
- 1992
(Show Context)
Citation Context ...plexities. In this abstract we only discuss the first four problems. For all these eight problems there are wt-optimal algorithms known on the CRCW-PRAM : linear work and constant time for Problems 1 =-=[15]-=- and 3 [12]; linear work and O(log log n) time for Problems 2 [6], 4 [12], 5 [4], 6 [13], 8 [4]; O(n log n) work and O(log log n) time for Problem 7 [3]. The\Omega\Gammae/4 log n) lower bounds for the... |

11 | An Optimal O(log log n) Time Parallel Algorithm for Detecting all Squares in a String
- Apostolico, Breslauer
- 1995
(Show Context)
Citation Context ...: linear work and constant time for Problems 1 [15] and 3 [12]; linear work and O(log log n) time for Problems 2 [6], 4 [12], 5 [4], 6 [13], 8 [4]; O(n log n) work and O(log log n) time for Problem 7 =-=[3]-=-. The\Omega\Gammae/4 log n) lower bounds for the time follow from [7]. The\Omega\Gamma n log n) lower bound for the work of Problem 7 follows from [23]. A logarithmic lower bound for all these problem... |

11 |
Deterministic sampling-a new technique for fast pattern matching
- Vishkin
- 1991
(Show Context)
Citation Context ...andidate elimination as well as the new one can be described in terms of warfare. The duels [27] constitute a medieval war (one comparison eliminates one or two candidates); the deterministic samples =-=[28]-=- can be viewed as a conventional war (a number of comparisons may eliminate a moderately large number of candidates); while the periodic mismatch can be viewed as a trigger for a nuclear winter (one c... |

10 |
Optimal parallel pattern matching
- Vishkin
- 1985
(Show Context)
Citation Context ...se problems on the CREW-PRAM follows immediately from the logarithmic lower bound for the AND function [9]. The only two cases where wt-optimal algorithms are known on the CREW-PRAM are for Problem 1 =-=[27]-=- and Problem 3 [2]. Surprisingly, very few (deterministic) wt-optimal algorithms are known on the more realistic models of parallel computation such as the EREW-PRAM and the hypercube. There exist wt-... |

9 |
Optimal EREW parallel algorithms for connectivity, ear decomposition, and stnumbering of planar graphs
- Gazit
- 1990
(Show Context)
Citation Context ...r broadcasting on the EREWPRAM does not leave much space for such algorithms. In fact, on the EREW-PRAM we are aware only of wtoptimal algorithms for sorting [11] and some algorithms on planar graphs =-=[19]-=-. On the hypercube the situation is even worse. Since even for such a basic problem as routing no wt-optimal algorithm is known, any problem with a wt-optimal algorithm on this model would be of speci... |

6 | Parallel detection of all palindromes in a string,” Theoretical Computer Science 141 - Apostolico, Breslauer, et al. - 1995 |

5 | Optimal parallel two dimensional witness computation - Cole, Galil, et al. - 1993 |

4 | Alphabet-independent optimal parallel search for three-dimensional patterns
- Karpinski, Rytter
- 1993
(Show Context)
Citation Context ...l PRAM models and the hypercube family for d dimensional pattern matching when d is fixed. The only known wt-optimal is for three dimensional text search given a preprocessed pattern on the CREW-PRAM =-=[21]-=-. No wt-optimal algorithm is known on the CRCW-PRAM. We believe that external witnesses, periodic mismatches and high dimensional pseudo-periods can be used, though they may not be sufficient. Recentl... |

2 |
Parallel two dimensional matching in log log n time
- Amir
- 1993
(Show Context)
Citation Context ...CREW-PRAM follows immediately from the logarithmic lower bound for the AND function [9]. The only two cases where wt-optimal algorithms are known on the CREW-PRAM are for Problem 1 [27] and Problem 3 =-=[2]-=-. Surprisingly, very few (deterministic) wt-optimal algorithms are known on the more realistic models of parallel computation such as the EREW-PRAM and the hypercube. There exist wt-optimal algorithms... |

1 |
Parallel two-dimensional pattern matching
- Galil, Park
- 1993
(Show Context)
Citation Context ...W-PRAM. The CRCW-PRAM algorithm is not in a form even close to the one needed for applying the techniques of Section 2 and we see no way to use it. An earlier simpler algorithm that is not wt-optimal =-=[17]-=- does not appear to be useful either. To obtain wt-optimal algorithms on the weaker models, we needed to design a new wt-optimal CRCW-PRAM algorithm. In Section 7 we sketch this new algorithm that use... |

1 |
Work-Time Optimal Parallel Prefix
- G��asieniec, Park
- 1994
(Show Context)
Citation Context ...hat external witnesses, periodic mismatches and high dimensional pseudo-periods can be used, though they may not be sufficient. Recently wt-optimal algorithms were designed for the following problems =-=[18]-=-: 9. prefix matching; i.e., finding for each position of the text the largest prefix of the pattern starting at it given a preprocessed pattern 10. computing smallest witnesses Both algorithms use lin... |