Pung: Unobservable communication over fully untrusted infrastructure
Pung is a (research) communication system where users can privately talk to each other over a fully untrusted channel. In particular, Pung provably hides all content and metadata in a conversation while withstanding global adversaries. See our paper for details about the protocol and its guarantees.
Pung is written in Rust and compiles under the nightly compiler (nighly is required for our benchmarking framework). If you wish to use the nightly compiler as default simply run:
$ rustup install nightly $ rustup default nightly
Alternatively, you can run the following from the Pung directory to set the nightly compiler only for Pung.
$ rustup install nightly $ rustup override set nightly
$ git submodule init $ git submodule update
Installing Cap'n Proto
$ cd deps/capnproto/c++ $ ./setup-autotools.sh $ autoreconf -i $ ./configure $ make check $ sudo make install
Installing XPIR's dependencies
XPIR depends on boost >= 1.55, gmp, and mpfr. You can install them as follows.
Ubuntu 14.04 (or later) / Debian:
$ sudo apt-get install libboost1.55-all-dev libmpfr-dev libgmp-dev
$ sudo pacman -S boost mpfr gmp
$ sudo emerge dev-libs/boost dev-libs/mpfr dev-libs/gmp
Note: there is no need to compile or install XPIR. It will be built and linked automatically when Pung is compiled (see below).
Compiling Pung's binaries
To compile Pung with debug symbols simply run:
$ cargo build. The resulting
binaries will be:
To compile PUng with compiler optimizations run:
$ cargo build --release. The resulting
binaries will be:
Running Pung's microbenchmarks
All our microbenchmarks are found in the
We use the [Criterion] (https://github.com/japaric/criterion.rs) library to measure
all running times and provide useful statistics.
To run all microbenchmarks, simply type:
$ cargo bench -- --test --nocapture bench.
The results will be found in the
To run a single (or a set of) microbenchmark, simply type:
$ cargo bench -- --test --nocapture [PREFIX].
Where [PREFIX] is a prefix of the name of the microbenchmark(s) that you wish to run. For instance,
since all our microbenchmarks start with "bench", using "bench" as the prefix runs all of them. However,
if one wished to only run the PIR microbenchmarks it is sufficient to pass in "bench_pir" as the prefix.
The first step is to launch the server (the -m flag tells it to expect 2 messages each round):
$ ./target/release/server -m 2
Pass in --help to see available options. The second step is to launch the client:
$ ./target/release/client -n "user1" -p "user2" -x "secret" -r 10 & $ ./target/release/client -n "user2" -p "user1" -x "secret" -r 10
The above will run two clients with ids "user1" and "user2" that will communicate with each other using secret "secret" for 10 rounds. One can also run a single client by having it talk to itself (i.e., passing the same argument to -n and -p for the client and setting the server to expect only one message: -m 1).
Pass in --help to see available options. It is important that the client and the server are run with the same options (e.g., retrieval type, optimization, number of buckets).
scripts for instructions on how
to run multiple clients and servers.