• Current software technology is unable to handle all these issues
• Book keeping is still done by the users
• Main directions of research
  • Design of concurrent algorithms
  • Design of concurrent languages
  • Construction of tools to do software development
    • Sequential compilers and book keeping tools
    • Parallelizing and vectoring compilers
    • Message passing libraries
  • Development of mathematical libraries
• Languages Fortran, C, Java, X10 etc.
• Dusty decks problem
  • Conversion of large body of existing sequential programs developed over last 40 years
  • Several billions lines of code and manual conversion is not possible
  • Restructuring compilers are required

Important Problems

Restructuring

• Identify parts of program to take advantage of characteristics of machine
• Manual coding is not possible
• Compilers are more efficient

Scheduling

• Exploit parallelism on a given machine
• Static scheduling: done at compile time
• Dynamic scheduling: done at run time
  • High overhead
  • Implemented through low level calls without involving OS (auto scheduling compilers)
• Auto scheduling compilers
  • Offer new environments for executing parallel programs
  • Small overhead
  • Parallel execution of fine grain granularity is possible
• Loop scheduling: Singly nested vs multiple nested
• Runtime overheads: Scheduling, synchronization, inter processor communication