| Term | Instructor |
|---|---|
| Fall 2017 | Tunga Güngör |
We are asked to implement MapReduce algorithm using Message Passing Interface (MPI) for calculating word frequencies. Master processor coordinates other slave processors. After this point I will use master for master processor and slaves for slave processors. Master shares data between slaves in balanced parts and combines results.
I used Boost MPI library for C++. It has more modern programatic interface than Open MPI
# Install Open MPI
brew install open-mpi
# Install Boost MPI. Please read any error message.
# It might be has linking error. You can google it to fix permission error.
brew install boost-mpi
# It just works with older versions of boots
brew install boost@1.60
brew link --overwrite --force boost@1.60
# Compile it
mpic++ -std=c++11 main.cpp -o main.o -I/usr/local/include/ -L/usr/local/lib -lboost_mpi -lboost_serialization
# Run it with 10 processors
mpirun --oversubscribe -np 10 main.o speech_tokenized.txt