KISTI-logoNational Supercomputing Center

KOR

Performance Optimization and Parallelization

Introduction

The supercomputing department provides scientific and technical guidance for users to implement and perform large scientific projects on the KISTI supercomputers. For that, we have annual program for performance optimization and parallelization of user code to achieve the required performance and solve large-scale problems. The program calls for collaboration to parallel and optimize the code for a year.

Program

  • Annual program

    • Calls for collaboration are made via the webpage "https://ksc.re.kr".
    • This targets parallel code development and optimization to solve large-scale problems.
    • In the case of writing a research paper using the code, co-authorship is required.
  • Users

    Current users of KISTI supercomputers or potential users who want to apply developed code on the supercomputer (※ No matter if you are a current user or not)

  • Condition
    • Users must use their own program.
    • The purpose of application is non-profit research, such as the publication of journals and so on.

    In research results (e.g., papers), an acknowledgment such as “The performance of the code was optimized (or parallelized) and tested with support from the KISTI National Supercomputing Center”must be included.

  • Expense

    The cost is free, but in the following cases,
    contact parallel@ksc.re.kr.

    • If the performance optimization/parallelization is not for research purposes.
    • If the code is not one’s own, such as open-source or commercial codes.
  • Call for proposal

    After registering as a member of this website, apply during the announcement period.

Regular proposal

It is planned to provide support for diverse methods of performance optimization to solve large-scale problems overall the current computational user code. This support will extend beyond short-term problem-solving for a minimum of 3 months up to a maximum of one year (with exceptions), and will include the following contents.

  • The announcement will be made through the website and e-mail. The application process will be announced on this website during the announcement period (scheduled for January). Applicants must apply on this website.
  • Single-node level optimization for expansion to large-scale problems, Traditional code optimization and parallelization, including OpenMP and MPI parallelization.
  • Algorithm change of existing code, parallelization technique development.
  • The development and application of code optimization techniques, including porting to a new programming environment and altering numerical algorithms for the application of accelerators (such as GPUs).
  • To improve the usability of highly influential community code, porting to supercomputers or other systems and improving performance.

Applicants cooperate with preparer (those who support the optimization/parallelization) on research and conduct projects. For these reasons, the Principal Investigator (PI) should propose the following details.

  • The explanation of the function and algorithms of the entire/each part of the code.
  • Providing an explanation of code usage and execution methods, along with a demonstration using examples.
  • Sharing of research results and recognition of rights after consultation.

Application / Review / Selection

Application steps

Sign in → Technology Support → Performance Optimization and Parallelization 새창으로 이동 Submit proposal.

Review and Selection

The review of the proposal contains the following consideration.

  • The suitability of usage on supercomputer
  • The feasibility of porting to a new programming environment
  • Application and impact of the code in community

Regular and long-term support will be provided for a minimum of 6 months to a maximum of 1 year (extendable), following the process.
The awarded projects that have been through written documents require an additional presentation. During this progress, the project’s direction and duration will be considered, and support will be provided afterward by the assigned preparator.

Contact