T. S. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components. In Proceedings of the 3rd Annual DIMACS Challenge, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....implemented on a coarsegrain MIMD machine. PRAM CC algorithms have been directly implemented only in the dataparallel computing model on SIMD machines [3] or on shared memory MIMD This research has been supported by MSMT Czech Republic under research program #J04 98:2123000 machines [4]. Any naive direct implementation on a MIMD machine with distributed memory is inefficient, because it results in a huge number of remote accesses. A distributed memory algorithm could be efficient if much of the work can be performed locally. For some classes of input graphs, such as graphs with ....

T. S. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components. In Proceedings of the 3rd Annual DIMACS Challenge, 1995.

....algorithms. In practice, programs get their random bits by using pseudo random number generators. Yet even in practice there are reports of algorithms giving quite different results under different pseudo random generators; see, e.g. FLW] for such reports on Monte Carlo simulations, and [Hsu, HRD] for the deviant performance of some RNC algorithms for graph problems. Other approaches involve using a physical source of randomness, such as a Zener diode, or using the last digits of a real time clock. Not only is it not clear that such Part of this work was done while the authors attended ....

T.-s. Hsu, V. Ramachandran, and N. Dean, Parallel Implementation of Algorithms for Finding Connected Components, Proc. DIMACS International Algorithm Implementation Challenge, 1994, pp. 1-14. 24

....get their random bits by using pseudo random number generators. Empirically, this often seems to be sufficient. However, there are reports of algorithms giving quite different results under different pseudo random generators (see e.g. FLW92] for such reports on Monte Carlo simulations, and [Hsu93, HRD94] for the deviant performance of some RNC algorithms for graph problems) An alternative approach is to use the output of some physical source of randomness, such 48 as a Zener diode, or the last digits of a real time clock. For such a source, it is plausible to assume that the string of bits ....

T.-s. Hsu, V. Ramachandran and N. Dean. Parallel implementation of algorithms for finding connected components. In Proc. DIMACS International Algorithm Implementation Challenge, pp. 1-14, 1994.

.... 1 1 Introduction and Motivations Although a significant amount of parallel machines have been built and a lot of parallel algorithms concerning graphs have been written, only a few implementations of those algorithms have been carried out on existing parallel platforms [HRD, KLCY94, KGP94, LKC95, RM94, HRD95] Being interested in graph problems, not only we would like to design parallel graph algorithms to process very large data as fast as possible, but also we would like that the implementations of these algorithms on parallel machines be as efficient as in ....

.... and Motivations Although a significant amount of parallel machines have been built and a lot of parallel algorithms concerning graphs have been written, only a few implementations of those algorithms have been carried out on existing parallel platforms [HRD, KLCY94, KGP94, LKC95, RM94, HRD95] Being interested in graph problems, not only we would like to design parallel graph algorithms to process very large data as fast as possible, but also we would like that the implementations of these algorithms on parallel machines be as efficient as in theory, and this for a vast scale of ....

[Article contains additional citation context not shown here]

T-S. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components in graphs. In 3rd DIMACS Challenge.

....for sparse random graphs of up to 8K nodes over the MasPar routines. However, the sequential machine did not have sufficient memory to store graphs larger than 8K nodes. While developing and testing these algorithms we created graphs on the fly as by some other research projects [KLCY94, HRD94] However, we found that this method presented two problems. First, duplicate edges were created which inflated the actual number of edges; our algorithms seemed to work faster. The second problem was that creating graphs on the fly precluded the possibility of accurately comparing algorithms ....

T. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components. In DIMACS implementation challenge, 1994.

.... high [17] There have also been reports of Monte Carlo simulations giving quite different results under different random number generators [12] and direct implementations of certain RNC algorithms taking longer time than expected due to the pseudorandom nature of computer generated random bits [15, 14]. In this paper, we present two new techniques for derandomization. The first leads to improved NC algorithms for many basic problems such as finding large cuts in graphs, set discrepancy, Delta 1) vertex coloring of graphs and others while the second improves the constructions due to [11] of ....

T.-s. Hsu, V. Ramachandran, and N. Dean, Parallel implementation of algorithms for finding connected components, in DIMACS International Algorithm Implementation Challenge, pages 1--14, 1994.

....get their random bits by using pseudo random number generators. Empirically, this often seems to be sufficient. However, there are reports of algorithms giving quite different results under different pseudo random generators (see e.g. FLW92] for such reports on Monte Carlo simulations, and [Hsu93, HRD94] for the deviant performance of some RNC algorithms for graph problems) An alternative approach is to use the output of some physical source of randomness, such as a Zener diode, or the last digits of a real time clock. For such a source, it is plausible to assume that the string of bits output ....

T.-s. Hsu, V. Ramachandran, and N. Dean, "Parallel implementation of algorithms for finding connected components," Proc. DIMACS International Algorithm Implementation Challenge, 1994, pp. 1--14.

.... arbitrary bandwidth to any memory location, but the inherent contention in the algorithm makes even EREW solutions much more challenging [6, 15, 18, 20] Implementation of the theoretical work has been restricted to shared memory machines [12] and SIMD machines with very slow processors [12, 17, 23]. Many practical solutions have been developed independently of theoretical work for modern MIMD massively parallel platforms (MPP s) 7, 10, 13, 14, 21, 26] and vector machines [9, 26] With the exception of [5] which focuses on 2D graphs for robot vision, these solutions typically emphasize ....

T.-S. Hsu, V. Ramachandran, N. Dean, "Parallel Implementation of Algorithms for Finding Connected Components," to be published in the Proceedings of the 3rd Annual DIMACS Challenge, 1995.

....constrained; it is a function of the random edge generation. Each vertex of a tertiary graph has degree 3. When no duplicate edges are allowed, a tertiary graph has 1:5n edges. 4.2. Generating Benchmark Graphs. Initially, we created graphs on the fly as other research projects had done [KLCY94, HRD94] However, we found that this method presented two problems. First, duplicate edges were created which inflated the actual number of edges, resulting in better performance for our algorithms. The second problem was that creating graphs on the fly precluded the possibility of accurately ....

T. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components. In DIMACS implementation challenge, 1994.

....for studying fundamental algorithmic issues. Many algorithms for more complex models are adaptations of algorithms first developed for a simple shared memory model. There are numerous examples, covering a wide range of problem domains, including sorting [18, 30, 44, 37] connected components [38, 42], computational geometry [66] FFT [22] and string matching [24] Designing an algorithm directly for the more complex model is typically a more daunting task than first developing the algorithmic insights on a simple shared memory model and only then adapting them to the more complex model. Note ....

T.-s. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components in graphs. In Proc. AMS/DIMACS Parallel Implementation Challenge Workshop III, 1997. To appear.

....and system routines. The performance data is shown in Table 8. Table 8 shows that our techniques for implementing virtual processing worked quite well on the code for graph algorithms. Details of our experiences in implementing parallel graph algorithms with virtual processing are given in [20,21]. 8 Concluding Remarks We have described techniques for implementing virtual processing on the MP1 using the MPL language. We have described our data allocation and code rewriting rules for writing MPL programs with virtual processing. We have also described the implementation of virtual ....

T.-s. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components. Presented at the 3rd DIMACS Implementation Challenge Workshop, 1994.

....for studying fundamental algorithmic issues. Many algorithms for more complex models are adaptations of algorithms first developed for a simple shared memory model. There are numerous examples, covering a wide range of problem domains, including sorting [17, 28, 42, 35] connected components [36, 40], computational geometry [61] FFT [21] and string matching [23] Designing an algorithm directly for the more complex model is typically a more daunting task than first developing the algorithmic insights on a simple shared memory model and only then adapting them to the more complex model. Note ....

T.-s. Hsu, V. Ramachandran, and N. Dean. Parallel implementation of algorithms for finding connected components in graphs. In Proc. AMS/DIMACS Parallel Implementation Challenge Workshop III, page to appear, 1997.

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