1: \documentclass[journal]{IEEEtran}%[times, 10pt,twocolumn]{article}

2: %\usepackage{latex8}

3: \usepackage{times,psfrag,subfigure,amssymb,amsmath,shortvrb,graphicx,alltt,comment}

4: %\usepackage{fullpage}

5: %\usepackage{amssymb,amsmath,shortvrb,graphicx,alltt}

6: %\textheight 8.75in \textwidth 6.5in \oddsidemargin 0cm

7: %\evensidemargin 0cm \topmargin 0in

8: \providecommand{\OO}[1]{\mathop{\mathrm{O}}\bigl(#1\bigr)}

9: \begin{document}

10: \def\reffig#1{Fig.\,\ref{#1}}

11: \def\reffeq#1{(\,\ref{#1})}

12: \def\reftable#1{Table.\,\ref{#1}}

13: \title{Lossy Bulk Synchronous Parallel Processing Model for Very Large Scale Grids}

14: \author{Elankovan Sundararajan, Aaron Harwood, Kotagiri Ramamohanarao

15: %

16: \thanks{E. Sundararajan, A. Harwood and Kotagiri Ramamohanarao

17: are with the Department of Computer Science and Software

18: Engineering, The University of Melbourne, ICT Building, 111 Barry

19: Street, Carlton 3053, Victoria Australia. Email: \{esund,aharwood,

20: rao\}@csse.unimelb.edu.au.}}

21: 

22: \markboth{IEEE TRANSACTION ON PARALLEL AND DISTRIBUTED SYSTEMS

23: ,~Vol.~-,No..~-,~Month~-}{Shell \MakeLowercase{\textit{et al.}}:

24: UDP-based Parallel Computing on Very Large Scale Grids}

25: 

26: 

27: \maketitle

28: 

29: \begin{abstract}

30: The performance of a parallel algorithm in a very large scale grid

31: is significantly influenced by the underlying Internet protocols

32: and inter-connectivity. Many grid programming platforms use TCP

33: due to its reliability, usually with some optimizations to reduce

34: its costs. However, TCP does not perform well in a high bandwidth

35: and high delay network environment. On the other hand, UDP is the

36: fastest protocol available because it omits connection setup

37: process, acknowledgments and retransmissions sacrificing reliable

38: transfer. Many new bulk data transfer schemes using UDP for data

39: transmission such as RBUDP, Tsunami, and SABUL have been

40: introduced and shown to have better performance compared to TCP.

41: In this paper, we consider the use of UDP and examine the

42: relationship between packet loss and speedup with respect to the

43: number of grid nodes. Our measurement suggests that packet loss

44: rates between $5\%$-$15\%$ on average are not uncommon between

45: PlanetLab nodes that are widely distributed over the Internet. We

46: show that transmitting multiple copies of same packet produces

47: higher speedup. We show the minimum number of packet duplication

48: required to maximize the possible speedup for a given number of

49: nodes using a BSP based model. Our work demonstrates that by using

50: an appropriate number of packet copies, we can increase

51: performance of parallel program.

52: \end{abstract}

53: 

54: 

55: \begin{keywords}

56: Modeling and prediction, probabilistic computation, parallelism,

57: UDP, performance analysis, parallel algorithm complexity.

58: \end{keywords}

59: 

60: \input{intro}

61: \input{approach}

62: \input{trans_los_pack}

63: \input{mpackcopies}

64: \input{algorithm}

65: \input{relwork}

66: \input{conc}

67: \section*{Acknowledgment}

68: % optional entry into table of contents (if used)

69: %\addcontentsline{toc}{section}{Acknowledgment}

70: Elankovan Sundararajan would like to thank The National University

71: of Malaysia and The University of Melbourne for providing

72: financial assistance.

73: 

74: \bibliographystyle{IEEEtran}%latex8}

75: \bibliography{IEEEabrv,UDP}

76: \end{document}

77: