WISETutorial

This project extends the capabilities of the WISE Toolkit to support a new class of experiments.

My final report contains a summary of the key changes I have made from the previous version of the WISE Toolkit.

Background & Tutorial

• An introduction to the WISE Toolkit & its uses can be found in proposal.pdf. Note, this toolkit was designed to support many of the experiments I proposed.
• A tutorial on how to run experiments & use the toolkit can be found here.
• Video explaining the differences between this toolkit and the previous generation of the toolkit. The slides found in this video can be found here.

Using this Toolkit

Running Experiments

Simply follow Appendix A of the tutorial. The steps where you allocate a public-facing IP via the OpenStack dashboard can be skipped since this has now been automated.

To run a Noisy Neighbor experimet, in the config file, add your stress testing machine(s). If you wish to run no stress-test, just leave replace "<FILL IN>" with "".

Note, also, that some of my scripts assume that you have an RSA public key/private key pair named elba.pub/elba that you use with your CloudLab account. Feel free to find-and-replace this out if necessary.

Make sure to install all the dependencies for the parsing and plotting scripts.

Files Modified in This Generation of the Toolkit

This is the list of files I had to edit to support the new features in this generation of the toolkit. It's not exhaustive -- it just highlights some key changes/contributions.

1. run.sh -- This is script used to launch the experiments and is therefore where we start the stress test. The stress-test is started in 428-444 and stopped in 558-581.
2. start_stress_test.sh file -- This file essentially configures the stress test.
3. openstack_setup.sh file -- This is where the public-facing IP is automatically assigned and where the VMs on OpenStack are launched. My contribution is contained in lines 37-41.
4. plot_graphs.py -- This is my script to generate a bunch of graphs. Note, it is a standalone file, and does not take any arguments; so feel free to read in whatever you need to to plot whatever you want. Currently, it just generates a very vanilla set of plots (many of which can also be ploted by the .gnuplot scripts in the /vis folder). However, I reccomend running this function only in the parse_results.sh since it depends on some folder creation that happens in this shell file.
5. parse_results.sh -- This file was created to automate the parsing and plotting of the data. Note, this file assumes that your results file is named results.tar.gz and is copied to experiment directory.
6. mem.py -- This is where the memory parsing error was fixed. In the percentage function, we save more digits when we round (line 15).
7. sample_data/ directory -- This folder contains sample data from two different experiments. The baseline experiment has no stress test running while the stress_test experiment does. View their respective results.tar.gz files and plots zip files.
8. compare_plots.py file -- This file allows you to plot the results from two different experiments side-by-side.

Configuring Noisy Neighbor Experiment

To simulate noisy neighbors, we run stress tests. You can stress whatever you want, as long as stress-ng supports it. To configure your test, simply edit the start_stress_test.sh file. You may also need to edit end_stress_test.sh, found in the same directory, based on what you're doing. If you find yourself constrained by stress-ng, in theory, you could also edit the start_stress_test.sh and stop_stress_test.sh files so that they install some other stress test package, starts it up, and stop it.

Other Things

• If you want to fork the repository and make changes, make sure to modify run.sh so that it clones the repository from your fork. Since we are cloning the repository inside this file without authentication, you'll probably need to make your copy of this repository public as well.
• If you want to run multiple different stress tests (e.g. one that stresses disk on one host, and one that stresses CPU on another), you can start by retracing my modifications in run.sh at the lines I mentioned.
• Feel free to reach out to me if there are any issues.
• In Appendix A of the tutorial, you can skip the section devoted to Public-Facing IP Allocation as this is automatically handled in this version of the tool.