Data and Results for my Research

Journal Data and Results Mentor Contact Me Home

View the rough draft of my final presentation here!

View the FAQ presentation here

View the report I wrote for the temperature-of-a-bar experiment here!

These graphs show all of the different machines' results. The top graph shows the running time in seconds for the simulation on each of the three sets of computers, and the bottom one shows the speedup (serial time / parallel time) for each of the three sets of computers. The simulations were run with 250 timesteps, 1000000 points on the bar, a timestep of .02, and a distance between points on the bar of .25. (NOTE: The reason the data jumps significantly at 15 processes for the 15-processor set is because that is the point where multiple processes are running on a single processor, which approximately doubles the running time.)

This is the running times for each of the simulations, using 1-30 parallel processes on each computer set.

This is the speedup (serial time / parallel time) for each of the simulations, using 1-30 parallel processes on each computer set.

The first graph with real data in it; it was run on a set of 15 Linux 64-bit machines with 3400MHz processors and 1 Gb of memory. The simulation was run with 250 timesteps, 500 points on the bar, a timestep of .02, and a distance between points on the bar of .25.

This simulation was run with double the number of points on the bar to try to minimize the overhead's effect on the data. It was run on a set of 15 Linux 64-bit machines with 3400MHz processors and 1 Gb of memory. The simulation was run with 250 timesteps, 1000 points on the bar, a timestep of .02, and a distance between points on the bar of .25.

This is a simulation run on Vega (64-bit Linux with 2 processors, 2133MHz, and 2Gb of memory) using 250 timesteps, 1000 points on the bar, a timestep of .02, and a distance between points on the bar of .25.

This is a simulation run on Culebra (64-bit Linux with 1 processor, 3400MHz, and 1Gb of memory) using 250 timesteps, 1000 points on the bar, a timestep of .02, and a distance between points on the bar of .25.

These are some preliminary graphs for my "toy" problem involving calculating temperature on a 1-D bar over time. Right now, I have only run the simulation on a maximum of two machines, with a maximum of 4 distinct processors.

Each simulation was run with 250 time steps, 5 points on the bar (counting both end points), a timestep of .02, and a distance between points on the bar of .25.

This is the simulation using 4 MPI processes run on Culebra, a 64-bit Linux machine with a single processor, with 3400MHz and 1Gb memory.

This is the simulation using 4 MPI processes run on my machine Vega, which is 64-bit Linux with 2 processors, 2133MHz, and 2Gb of memory.

This is the simulation using 4 MPI processes run on Vega and Carstensz, both 64-bit Linux machines with 2 processors each (total of 4 processors). Vega has 2133MHz and 2Gb of memory and Carstensz has 1867MHz and 2Gb of memory.