• About us
  • Contact us
  • Sitemap
  • Home
  • About us
    • Mission Statement
    • Management and Governance
      • Governance Committee
      • Management Committee
      • Industry Advisory Board
      • Scientific Advisory Board
    • Lero Contact Information
    • Institutions
    • Contact Us
  • Research
    • Competencies
    • Projects
    • Posters
  • Industry
    • Industry Outreach
    • Collaborating with Lero
    • Partners
    • Intellectual Property
  • Education and Outreach
    • Second Level
      • Internships
      • School Visits
      • Scratch Lesson Plans
      • UL Cybercamp
    • Third Level
    • Fourth Level
      • Doctoral Symposium
      • Writers Retreat
      • LGSSE
    • Past Activities
  • Publications
    • PhD Thesis
    • Publication List
    • Technical Reports
  • Events
    • Upcoming Events
    • Events Calendar
    • Events Map
    • Conference Information
    • Conference List
    • List of Talks
  • News
    • Newsletters
  • Partners
  • People
    • Senior Academic Team
    • Staff Directory
    • Directors
    • Vacancies
    • Alumni
    • Visitors to Lero
      • Past Visitors

Menu

  • Home
  • About us
  • Research
  • People
  • Partners
  • Education and Outreach
  • Events
  • Industry
  • Publications
  • News
  • Contact us
Home | MANYCORE

Project MANYCORE

Understanding Software Performance on Many-Core Systems

In recent years almost all processor manufacturers have switched from building processors with a single processor core to providing multiple cores on the same chip. Developing parallel software is difficult, time consuming and expensive. Solving the multi-core software problem is one of the most important challenges facing software engineering today. A particularly difficult problem on many-core systems is understanding software performance. As the number of cores rises, the interactions between parallel threads, and the interactions between the threads and the hardware become increasingly important for software performance. However, understanding these complex interactions is a difficult problem for ordinary software developers. The MANYCORE work programme will address this problem by developing tools and techniques for understanding software and thread interactions on many-core systems.

Understanding software performance has always been a problem for software developers. Interactions between processor pipelines and caches can cause behaviour which is difficult to understand even on machines with only a single processor core. However, with multiple processor cores, the scope for complex interactions increases significantly. For example, four threads might be running on separate cores, but may affect one another’s performance by contending for shared caches and memory bandwidth, so the behaviour of each thread affects all others. On simultaneous multi-threaded processors, where multiple threads run in parallel on the same core, the interactions are even more complex, because different threads may contend for the same execution resources. However, when we consider future many-core processors with perhaps 64 parallel threads interacting in complex ways, understanding performance becomes almost impossible with existing techniques, even for software performance experts. The goal of this work programme is to create software development tools that allow ordinary developers to understand the performance of software on many-core systems.

The work programme is in collaboration with IBM Research Lab Dublin, and will use IBM PowerEN technology.

Project Leader
David Gregg
Project Team
Roman Atachiants
Partners
Related Competencies: 
Embedded Systems
Parallel and Multicore Computing
Performance Engineering
Printer-friendly versionPrinter-friendly version

Lero - The Irish Software Engineering Research Centre; Tel: +353 61 233799; Fax: +353 61 213036; Contact us