Skip to content

Software artifacts for the article "Multiparty Homomorphic Encryption from Ring-Learning-with-Errors" presented at the 21st Privacy Enhancing Technologies Symposium (PETS'21)

License

Notifications You must be signed in to change notification settings

ldsec/lattigo-pets21

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Multiparty Homomorphic Encyption from Ring-Learning-with-Errors: Artifacts

This repository hosts the software artifacts for the article Multiparty Homomorphic Encryption from Ring-Learning-with-Errors [1] presented at the 21st Privacy Enhancing Technologies Symposium (PETS'21).

Artifacts list

The following sowftare items are artifacts of the article:

Artifact Description
lattigo/v2/dbfv the Lattigo Go package implementing the multiparty BFV scheme.
lattigo-pets21/apps a Go module importing the github.com/ldsec/lattigo/v2/dbfv Lattigo sub-package and implementing the experiments
lattigo-pets21/apps/pir a Go application that implements the PIR experiment
lattigo-pets21/apps/psi a Go application that implements the PSI experiment
lattigo-pets21/apps/tpl a Go application that implements the Beaver-triples-generation experiment

The lattigo/v2/dbfv package is integrated in the official Lattigo repository at https://github.com/ldsec/lattigo. The lattigo-pets21/apps module is in this repository and imports the latest version of the lattigo/v2/dbfv package via a Go module dependency. This repository includes a Makefile, a Dockerfile and several scripts that automate building and running our code.

Building

From a clone of this repository on a machine running Docker, running

make

will build a mhe-exps docker image for which the three experiment apps' binaries are in the PATH.

Running

Multiparty-Input-Selection (PIR) and Element-Wise-Vector-Product (PSI) experiments

The PIR and PSI experiments are local and running the client and server within the same process. Both programs take the number of input-parties and the number of goroutines (threads) for the circuit-evaluation by the cloud:

docker run --rm mhe-exps [psi|pir] [#parties] [#goroutines] 

Exemples:

docker run --rm mhe-exps pir        # runs the PIR experiment over 8 parties with a single-threaded cloud evaluation

docker run --rm mhe-exps psi 16     # runs the PSI experiment over 16 parties with a single-threaded cloud evaluation

docker run --rm mhe-exps psi 16 8   # runs the PSI experiment over 16 parties with cloud evaluation using 8 threads

Multiplication-Triple-Generation experiment

The Beaver-triples-generation experiment runs every party in its own process, by running several instances of the mhe-exps image within docker network named mpc-net. The run-tpl-parties.sh script automates the process of starting the experiment for a given generation technique and number of parties.

./run-tpl-parties.sh [he|mhe] [#parties] [filename]

The stdout of party 0 is redirected to the host stdout. The script also accepts a filename as an option final argument. If provided, it saves the stdout of each party to a file [filename]_p[party id].txt.

Finally, the run-tpl-exp.sh automates the process of running the Beaver-triples-generation experiment for both the he and mhe generation techniques, for 2 to 8 parties. The stdout of each party in each experiment is redirected to a file in the output directory.

Note: Dockerization of the experiment seems to be a little less stable than our initial setting, especially when run on less powerful systems. Some isolated experiments might fail because docker cannot bring the container up fast enough and some tcp connections are sometime reset. These experiments can be restarted indivitually by using the run-tpl-parties.sh script with the corresponding arguments.

Cleaning up

There are two make targets for the clean-up tasks:

make clean-output: deletes the output directory.

make clean-docker: removes the mhe-exps images from the Docker host and the mpc-net docker network.

make clean-all: performes the clean-output and clean-docker targets.

References

[1] Christian Mouchet, Juan Troncoso-Pastoriza, Jean-Philippe Bossuat, Jean-Pierre Hubaux. 2021. Multiparty Homomorphic Encryption from Ring-Learning-with-Errors. To be presented at PETS'21.

About

Software artifacts for the article "Multiparty Homomorphic Encryption from Ring-Learning-with-Errors" presented at the 21st Privacy Enhancing Technologies Symposium (PETS'21)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages