CONTENTS
CONTENTS
CONTENTS
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
230 VIRGINIA NICULESCU (1)<br />
4. Conclusions<br />
One-dimensional, multidimensional, and set-distributions could be defined for<br />
PowerList and PowerArray data structures, and functions defined on these special<br />
data structures are transformed to accept distributed input data, and then costs<br />
analysis could be done.<br />
We have several advantages of formally introducing the distributions; the first is<br />
that it allows us to evaluate costs, depending on the number of available processors -<br />
as a parameter. In the PowerArray case, this is especially important, since we may<br />
choose the most advantageous factorization of the number of processors on each dimension.<br />
The analysis of the possible distributions for a certain function may lead to<br />
an improvement in the design decisions, too. Another advantage is that we may control<br />
the parallel decomposition until a certain level of tree decomposition is achieved;<br />
otherwise parallel decomposition could be done, for example, in a ‘deep-first’ manner,<br />
which could be disadvantageous. Also, after the introduction of the distributions<br />
functions, mapping on real architectures with limited number of processing elements<br />
(e.g. hypercubes) could be analyzed.<br />
ACKNOWLEDGEMENT<br />
This work was supported by the research project ID 2268/2009, sponsored by the<br />
Romanian National University Research Council (CNCSIS).<br />
References<br />
[1] Gorlatch, S.,: Abstraction and Performance in the Design of Parallel Programs, CMPP’98<br />
First International Workshop on Constructive Methods for Parallel Programming, 1998.<br />
[2] Kornerup, J.: Data Structures for Parallel Recursion. PhD Thesis, Univ. of Texas, 1997.<br />
[3] Misra, J.: PowerList: A structure for parallel recursion. ACM Transactions on Programming<br />
Languages and Systems, Vol. 16 No.6 (1994) 1737-1767.<br />
[4] Niculescu, V.: On Data Distributions in the Construction of Parallel Programs, The<br />
Journal of Supercomputing, Kluwer Academic Publishers, 29(1): 5-25, 2004.<br />
[5] Niculescu, V.: “Data Distributions in PowerList Theory”, Lecture Notes of Computer<br />
Science Vol. 3722: Theoretical Aspects of Computing, Proceedings of ICTAC 2007, Springer-<br />
Verlag, 2007: 396-409.<br />
(1) Department of Computer-Science, Babes-Bolyai University, 1 M. Kogalniceanu,<br />
Cluj-Napoca, Romania<br />
E-mail address: vniculescu@cs.ubbcluj.ro