No description
Find a file
2019-05-28 17:40:49 +02:00
scripts Fix pushd/popd code 2019-05-23 11:03:25 +02:00
src Avoid looking at headers 2019-05-28 17:40:49 +02:00
.gitignore Add a gitignore 2019-02-21 14:27:09 +01:00
CMakeLists.txt Add capdiff to see differences between captures 2019-05-28 17:15:37 +02:00
Dockerfile Upgrade build stack 2019-05-15 15:52:39 +02:00
README.md Improve 2019-05-23 22:36:26 +02:00
torrc_simple Less log, fix buffer packet management 2019-02-20 09:34:10 +01:00

Donar

Quickstart

Installation

The following steps are provided for Fedora 29 Workstation. We assume you have two containers or two virtual machines or two physical machines.

To setup each machine, you should do:

sudo dnf install --refresh -y cmake gcc gcc-c++ ninja-build glib2-devel glib2 tor valgrind git net-tools nmap-ncat
git clone https://gitlab.inria.fr/qdufour/donar.git
cd donar
mkdir out && cd out && cmake -GNinja .. && ninja && sudo ninja install

Commands

Now your machine is ready and you will be able to use the following commads:

  • donar is our main binary. It can be run as a client or a server.
  • udpecho is a simple udp server that send back the data sent to him.
  • torecho is a simple tcp server that send back the data sent to him + configure the tor daemon to generate a hidden service URL and be accessible on the Tor network.
  • measlat can be used in conjunction with udpecho or torecho to measure a Round Trip Time (RTT)

Try to run the previous commands in your terminal without any option, you will see their help.

At any moment, you can use the following commands that are not part of the project to understand what you are doing:

netstat -ulpn # Show programs that listen on an UDP port
netstat -tlpn # Show prograns that listen on a TCP port
nc 127.0.0.1 8000 # Connect via TCP on server 127.0.0.1 listening on port 8000
nc -u 127.0.0.1 8000 # Connect via UDP on server 127.0.0.1 listening on port 8000

Introduction to the debug tools udpecho and measlat

Now let's start simple, we will launch our udp echo server and access it locally:

udpecho -p 8000 &
nc 127.0.0.1 8000

If you write some data on your terminal and press enter, you will see that your data has been repeated. Well done!

Now, instead of using nc, we will use measlat to use this echo server to measure latencies (make sure that udpecho is still running):

measlat -h 127.0.0.1 -p 8000 -t udp

measlat will send one packet to our udpecho server and wait to receive it back, measure the time it took, display it and exit.

You can use measlat more extensively by defining the number of measures to do, an interval and the size of the packet:

measlat -h 127.0.0.1 -p 8000 -t udp -c 10 -i 100 -s 150

Introduction to donar

Now, let's introduce our main project. First, kill all the remaining processes killall udpecho measlat nc.

On both machine
Move to the donar repository root where you will see the torrc_simple file. We will need to start by launching tor in a terminal:

tor -f ./torrc_simple

On machine A
Launch Donar server in a second terminal:

donar -s -a naive -e 3000 -r 3001

In a third terminal, launch your echo service:

udpecho -p 3000

Display the content of the file onion_services.pub that has been created in your working directory.

On machine B
Copy the content of the file onion_services.pub that is on machine A to machine B in a file named machine_a.pub. Now, run Donar client in a second terminal:

donar -c -a naive -o machine_a.pub -r 3000 -e 3001

In a third terminal, launch your echo service:

udpecho -p 3001

On machine A You can access to the echo service from machine B by running:

nc 127.13.3.7 3001
# or
measlat -h 127.13.3.7 -p 3001 -t udp

On machine B You can access to the echo service from machine A by running:

nc 127.13.3.7 3000
# or
measlat -h 127.13.3.7 -p 3000 -t udp

If it works, that's all! You are now mastering Donar!

Linphone configuration

Choose a SIP UDP, Audio RTP/UDP and Video RTP/UDP that is different between your clients. Go to manage account. Add a new SIP proxy.

Username: <username>@127.13.3.7:5061
Proxy: 127.13.3.7:5060
Leave the rest empty.
Uncheck all the checkboxes.

You also need to say to Linphone that you are behind a NAT and put 127.13.3.7 as your public IP address.

Docker build

sudo docker build -t registry.gitlab.inria.fr/qdufour/donar .
sudo docker push registry.gitlab.inria.fr/qdufour/donar
sudo docker pull registry.gitlab.inria.fr/qdufour/donar
mkdir -p ./{xp1-shared,xp1-res}
sudo chown -R 1000 ./{xp1-shared,xp1-res}

sudo docker run -t -i \
  --privileged \
  -v `pwd`/xp1-shared:/home/donar/shared \
  registry.gitlab.inria.fr/qdufour/donar \
  xp1-server
 
sudo docker run -t -i \
  --privileged \
  -v `pwd`/xp1-res:/home/donar/res \
  -v `pwd`/xp1-shared:/home/donar/shared \
  registry.gitlab.inria.fr/qdufour/donar \
  xp1-client 1000 100 100

Run an XP instance

sudo ./scripts/xp1 1000 100 100

Run instances in parallel

We generate the name of the algorithm to run on the right side of the parallel command. The idea is to generate a sequence like the following: orig naive rr rrh orig naive rr rrh orig....

parallel -j 12 bash -c './xp-instance-runner $1 6000 100 100' -- `xp0=orig xp1=naive xp2=rr xp3=rrh xp4=witness; for i in {0..99}; do q='xp'$((i % 5)); echo ${!q}; done`
parallel.moreutils -j 16 bash -c './xp-instance-runner $0 6000 100 100' -- `xp0=orig xp1=naive xp2=rr xp3=rrh xp4=witness; for i in {0..274}; do q='xp'$((i % 5)); echo ${!q}; done`
parallel.moreutils -j 16 bash -c './xp-instance-runner $0 6000 100 100' -- `for i in {0..55}; do echo -n 'orig naive rr rrh witness '; done`

Tests:

parallel.moreutils -j 16 bash -c './xp-instance-runner rr 6000 100 100' -- `seq 0 15`

Helpers

Track measures that didn't finish:

ls | grep -P '^naive-|^rrh-|^rr-|^orig-' | while read n; do echo -n "$n "; tail -n1 $n/res/*.csv ; done | grep -v "Measurement done"

Check if timer's bug occured:

ls | grep -P '^naive-|^rrh-|^rr-|^orig-' | while read n; do echo "$n "; grep 'bug' $n/log/client-measlat-stderr.log; done

Check if a free() invalid pointer bug occured:

grep 'free' naive-*/log/*-donar-*.log
grep -rn 'free()' .