CPU Benchmark

Scenario: You want to test which CPU is better between some Linux machines.

• Download the (small) repository mini-nbody:

git clone https://github.com/harrism/mini-nbody

or alternately, copy the nbody.c and and timer.h files to the servers you want to test, inside e.g. ~/nbody-mini directory.

• Add the following line at the end of the main() function of the nbody.c file:

printf("Total time: %0.3f secs\n", totalTime);

Important

If you don’t want this benchmark to stress all the CPUs - which is enabled by default (e.g. a machine with 8 cores will show 800% in top command if you run this program) you MUST COMMENT this line that uses the OpenMP standard as is below:

//  #pragma omp parallel for schedule(dynamic)

Thus, with the above comment the nbody program will reach 100% (one core only full utilization) and will not bring a production system to its knees by using all the available CPUs!

• Execute:

SRC=nbody.c
EXE=nbody
gcc -std=c99 -O3 -fopenmp -DSHMOO -o $EXE $SRC -lm -D_DEFAULT_SOURCE
./nbody

Note

• By default nbody.c takes as an argument nbodies=30000, so if you want to test for more time, use it as such: ./nbody 40000 (the larger the value of nbodies, more calculations are needed, so more CPU time)
• Be careful to compare the same number of nbodies between different servers with different CPUs.
• Less totalTime for one Core means better CPU (so you can safely say which server is the faster one :)