Lesson 12 Date: 4/16/2014

High Performance Computing (part I)

Linux for Engineering and IT applications

Point to Point Communication In two communicating MPI processes: one task is performing a send operation and the other task is performing a matching receive operation. MPI_Send(buffer,count,type,dest,tag,comm) ---> MPI_Recv(buffer,count,type,source,tag,comm,status) A System Buffer allows asynchronous send-receive operations. User managed address space (i.e. your program variables) is called the application buffer. MPI Send/Receive can be blocking and non-blocking. Blocking - the function won't return until MPI has safely stored the message to be sent, or was able to retrieve a message for delivery.

Parallel Computing HPC paradigms Amdahl's law OpenMP Parallel regions and loops Parallel sections Synchronization Constructs Message Passing Interface (MPI) General MPI Program Structure Point to Point Communication MPI Standard Data Types MPI Send/Receive example MPI Collective Communications Collective Communication Example Master-Slave Task Farming Computing Partial Sums VM for MPI cluster OpenMPI installation MPI code compilation and run in OpenMPI References Take me to the Course Website