Efficient Use of Compute Resources

1. Reasons for Wasted Cores

• Not enough Threads
• Blocked Threads lead to not enough threads

2. Goal

• need at least one ready/running thread for each CPU core
• generally need more threads than cores (threads are often blocked)
• threads could be in one process (or many)

3. Solutions

• Multi-threaded applications
  • good when multiple threads need to access frequently modified data structures
  • new kinds of bugs possible (race conditions, deadlock)
• Multi-process applications
  • easier to program (or just manually launch several processes in background)
  • better at keeping multiple cores busy simultaneously (Python specific)

Virtual Address Spaces

1. Process and Address Spaces

• A process is a running program, runs independently and has its own memory space, resources, and system resources.
• Virtual Address:
  • Regular processes cannot directly access physical memory or other address spaces (avoid spying)
  • Each process has its own virtual address space
  • The same virtual address generally refers to different memory in different processes
  • Virtual address spaces can have “holes” (the length of virtual address is usually greater than physical address)
  • Physical memory for a process need not be contiguous
• Address Space in Details:

  • Locations of Code and Variables:

    

  • CPU Execution

    • CPUs are attached to at most one instruction pointer at any given time (CPU points to interpreter code, which points to Python bytecode)
    • They run code by executing instructions and advancing the instruction pointer

Threads