OpenMP + MPI 编程 （七）Quiz

1. Hybrid programming combines what two models of computing?
   Shared-memory parallel and distributed-memory parallel
   Parallel and serial
   Message-passing and NUMA
   All of the above
   Correct.

2. An MPI process is generally single-threaded unless the code has been augmented with multithreading directives or library calls.
   True
   False
   Correct.

3. What is a typical way for a batch script to assign MPI processes to the hardware, for a hybrid program?
   One MPI process per node
   One MPI process per socket
   One MPI process per core
   Either (a) or (b)
   Correct.

4. At what point does one invoke numactl on a process?
   Dynamically, after a process is running
   At the start of process execution
   Within the program, through an F90/C API
   Any of the above
   Correct.

5. The -i option to numactl is used to:
   Make interleaved (round-robin) memory allocations among available sockets
   Make an individual assignment of a process, so it has affinity to just the specified core
   Make an intra-socket assignment of a process, so it has affinity to any core on the specified socket
   None of the above
   Correct.

6. If MPI_Init_thread returns MPI_THREAD_FUNNELED, MPI messages can only be passed between main threads.
   True
   False
   Correct.

7. If MPI_Init_thread returns MPI_THREAD_SERIALIZED, MPI messages can be passed between any two threads, provided each is enclosed in an “omp single” construct.
   True
   False
   Correct.

8. Other threads of a given process can do no work while one thread is using MPI to communicate.
   True
   False
   Correct.

9. Which argument to MPI_Send may be used to identify the destination thread?
   rank
   count
   tag
   communicator
   Correct.

10. At what point does one invoke sched_setaffinity on a thread?
    Dynamically, after a process is running
    At the start of process execution
    Within the program, through an F90/C API
    Any of the above
    Correct.