Readme for the VM/Container boot time experiments

The experiments on VM/Container boot time are conducted by using the scripts extended from vm5k (see below).

The scripts can be found in engines/ with:

• VMBootTime.py: experiments on boot time of VMs
• ContainerBootTime.py: experiments on boot time of Containers

Each script takes several command-line parameters as input, you can use: ./VMBootTime.py -h or ./ContainerBootTime.py -h to print the list of parameters needed.

In order to run the script, you need access to Grid5000 servers. The scripts can be modified to be used in other platforms as well. After git clone the repository and choose the experiment to perform, you can run the experiment with your custom scenario. For example:

./VMBootTime.py --vm 16 econome -c econome-test —monitor

will run the VM boot time experiment with 16 VMs on the cluster named econome and the output is saved in folder econome-test. The output will be the boot time of VMs and the resource usage.

BootTime2csv.py is the script to convert the results of the experiments, which can be found in postprocess/. To run the script (use argument -h to see the list of all other arguments):

python boottime2csv.py -i <input_directory> -o results.csv

The csv file contains information about the experiments, each row is one run of a experiment with the columns are the scenario parameters and the boot time result. You can do some further analysis on this file using RStudio, pandas, MS Excel or the tool of your choice.

We use this image for setting up VMs in those experiments. This is a Debian 7 image, in qcow2 format. The image has all the benchmarks we used in our experiments, including:

• LINPACK is used to produce CPU workloads. LINPACK estimates a system's floating point computing power by measuring how fast a computer solves a dense n by n system of linear equations Ax = b.
• CacheBench is a benchmark to evaluate the raw bandwidth in megabytes per second of the memory of computer systems. It includes read, write and modify operations on the memory to fully simulate the effect of memory usage.
• Iperf measures the maximum achievable bandwidth on IP networks. Iperf creates TCP and UDP data streams to measure the throughput of a network that is carrying them.
• Stress simulates an I/O stress by spawning a number of workers to continuously write to files and unlink them.

You can find the results for analyzing VMs boot time in here and Containers boot time in here.

A python module to ease the experimentations of virtual Machines on the Grid'5000 platform. It is composed of:

• a script that deploy virtual machines (vm5k)
• an experimental engine that conduct user defined workflow for a set of parameters (vm5k_engine)
• a lib to setup Debian hosts with libvirt and manage virtual machines

Developed by the Inria Hemera initiative 2010-2014 https://www.grid5000.fr/mediawiki/index.php/Hemera

See documentation on http://vm5k.readthedocs.org

The module requires:

• execo 2.4, http://execo.gforge.inria.fr/

Connect on a Grid'5000 frontend and type the following commands::

export http_proxy="http://proxy:3128" export https_proxy="https://proxy:3128" easy_install --user execo easy_install --user vm5k

Add .local/bin to your path and run vm5k !

• Laurent Pouilloux
• Daniel Balouek-Thomert
• Jonathan Rouzaud-Cornabas
• Flavien Quesnel
• Jonathan Pastor
• Takahiro Hirofuchi
• Adrien Lèbre

• Matthieu Imbert
• Simon Delamare