본문 바로가기
  • Home

A Study on Determination of the Number of Work Processes Reflecting Characteristics of Program on Computational Grid

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2006, 11(1), pp.71-86
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science

조수현 1 Kim, Young-Hak 1

1금오공과대학교

Candidate

ABSTRACT

The environment of computational grid is composed of the LAN/WAN each of which has different efficiency and heterogeneous network conditions, and where various programs are running. In this environment, the role of the resource selection broker is very important because the work of each node is performed by considering heterogeneous network environment and the computing power of each node according to the characteristics of a program. In this paper, a new resource selection broker is presented that decides the number of work processes to be allocated at each node by considering network state information and the performance of each node according to the characteristics of a program in the environment of computational grid. The proposed resource selection broker has three steps as follows. First, the performance ratio of each node is computed using latency-bandwidth-cpu mixture information reflecting the characteristics of a program, and the number of work processes that will be performed at each node are decided by this ratio. Second, RSL file is automatically made based on the number of work processes decided at the previous step. Finally, each node creates work processes by using that RSL file and performs the work which has been allocated to itself. As experimental results, the proposed method reflecting characteristics of a program, compared with the existing (uniformity) and latency-bandwidth method is improved 278%~316%, 524%~595%, 924% ~954% in the point of work amount, work process number, and node number respectively.

Citation status

* References for papers published after 2023 are currently being built.