Skip to content

README.md

Latest commit

 

History

History
130 lines (94 loc) · 3.67 KB

README.md

File metadata and controls

130 lines (94 loc) · 3.67 KB
Raw

MessageFirst: a reliable user API for message-oriented network workloads

This repository contains code for my thesis project where I designed and implemented MessageFirst, which is a user API for message-oriented network workloads. It is a blocking API that supports the notion of a "message", a unit of transmission that is not considered done until a message has been sent and a respective response has been received.

Downloading, building and running tests

Before running any benchmarks, first build MessageFirst and run the tests to make sure your environment is set up correctly.

Downloading MessageFirst

Clone the repo as so:

git clone https://github.com/jakemassoth/MessageFirst.git
cd MessageFirst

Python

We use Miniconda to create isolated python environments with the necessary packages and versions needed to run the MessageFirst tests and benchmarks.

Download the Miniconda installer for Linux:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
chmod +x ./Miniconda3-latest-Linux-x86_64.sh
./Miniconda3-latest-Linux-x86_64.sh

The installer will ask where Miniconda should be installed. If you are running the benchmarks locally, feel free to use the default. If you are running the benchmarks on the DAS-5, make sure to install Miniconda in /var/scratch/<USERNAME>/miniconda3.

Answer "yes" when the installer asks if it should run conda init.

Now that Miniconda is installed, we can create the MessageFirst Python environment.

source ~/.bashrc
conda create --name MessageFirst --file spec-file.txt
echo "conda activate MessageFirst" >> ~/.bashrc
source ~/.bashrc
rm Miniconda3-latest-Linux-x86_64.sh

You should now see (MessageFirst) prepended to your terminal prompt.

Building MessageFirst

Requires CMake 3.15.

cmake .
make
make test

If these all run successfully, you're good to go!

Running Examples

TODO

Running Benchmarks

This repository contains two benchmarks:

  • netperf_stream_rr compares the throughput achieved by netperf's TCP_RR test to a baseline TCP_STREAM test
  • echo_benchmark benchmarks the MessageFirst API implementation.

Installing netperf

First, install netperf for your distribution. Netperf is a network benchmarking tool used in the netperf_stream_rr benchmark. On Ubuntu 18.04:

sudo apt update
sudo apt install netperf

On the DAS-5:

module load netperf
module load cmake/3.15.4
module load gcc

Verify that netperf is working with a quick test benchmark:

netserver
netperf

This benchmark should complete successfully after about 10 seconds and display some statistics.

Running locally

Running benchmarks locally is pretty simple:

python benchmarks/run_benchmarks.py

This will take a few minutes to run. It will generate .csv files containing the data from the benchmarks in the relevant data folders for each benchmark.

To plot graphs:

python benchmarks/make_plots.py

This will generate .png plots for each benchmark in its relevant plots folder.

Running on DAS-5

First, reserve two nodes as so:

preserve -np 2 -t 900

Use preserve -llist to list the reservations currently active. Note the node numbers you have been assigned.

Then run the benchmark as so:

python benchmarks/run_benchmarks.py node<X> node<Y>

Replace <X> and <Y> with the node numbers of your reservation.

After this completes (it should take a few minutes), make the plots:

python benchmarks/make_plots.py

This will generate .png plots for each benchmark in its relevant plots folder.