A fast and portable C++11 library for Oblivious Transfer extension (OTe). The primary design goal of this library to obtain high performance while being easy to use. This library currently implements:
- The semi-honest 1-out-of-2 OT [IKNP03].
- The semi-honest 1-out-of-2 Delta-OT [IKNP03],[BLNNOOSS15].
- The semi-honest 1-out-of-N OT [KKRT16].
- The malicious secure 1-out-of-2 OT [KOS15].
- The malicious secure 1-out-of-2 Delta-OT [KOS15],[BLNNOOSS15].
- The malicious secure 1-out-of-N OT [OOS16].
- The malicious secure approximate K-out-of-N OT [RR16].
- The malicious secure 1-out-of-2 base OT [NP01].
- The malicious secure 1-out-of-2 base OT [CO15] (Faster Linux ASM version disabled by default).
This library provides several different classes of OT protocols. First is the base OT protocol of Naor Pinkas [NP01]. This protocol bootstraps all the other OT extension protocols. Within the OT extension protocols, we have 1-out-of-2, 1-out-of-N and K-out-of-N, both in the semi-honest and malicious settings.
All implementations are highly optimized using fast SSE instructions and vectorization to obtain optimal performance both in the single and multi-threaded setting. See the Performance section for a comparison between protocols and to other libraries.
Networking can be performed using both the sockets provided by the library and external socket classes. See the networking tutorial for an example.
A minimal working example showing how to perform n
OTs using the IKNP protocol.
void minimal()
{
// Setup networking. See cryptoTools\frontend_cryptoTools\Tutorials\Network.cpp
IOService ios;
Channel senderChl = Session(ios, "localhost:1212", SessionMode::Server).addChannel();
Channel recverChl = Session(ios, "localhost:1212", SessionMode::Client).addChannel();
// The number of OTs.
int n = 100;
// The code to be run by the OT receiver.
auto recverThread = std::thread([&]() {
PRNG prng(sysRandomSeed());
IknpOtExtReceiver recver;
recver.genBaseOts(prng, recverChl);
// Choose which messages should be received.
BitVector choices(n);
choices[0] = 1;
//...
// Receive the messages
std::vector<block> messages(n);
recver.receiveChosen(choices, messages, prng, recverChl);
// messages[i] = sendMessages[i][choices[i]];
});
PRNG prng(sysRandomSeed());
IknpOtExtSender sender;
sender.genBaseOts(prng, senderChl);
// Choose which messages should be sent.
std::vector<std::array<block, 2>> sendMessages(n);
sendMessages[0] = { toBlock(54), toBlock(33) };
//...
// Send the messages.
sender.sendChosen(sendMessages, prng, senderChl);
recverThread.join();
}
The running time in seconds for computing n=224 OTs on a single Intel
Xeon server (2 36-cores Intel Xeon CPU E5-2699 v3 @ 2.30GHz and 256GB of RAM
)
as of 11/16/2016. All timings shown reflect a "single" thread per party, with the
expection that network IO in libOTe is performed in the background by a separate thread.
Type | Security | Protocol | libOTe (SHA1/AES) | Encrypto Group (SHA256) | Apricot (AES-hash) | OOS16 (blake2) | emp-toolkit (AES-hash) |
---|---|---|---|---|---|---|---|
1-out-of-N (N=276) | malicious | OOS16 | 10.6 / 9.2 | ~ | ~ | 24** | ~ |
1-out-of-N (N=2128) | passive | KKRT16 | 9.2 / 6.7 | ~ | ~ | ~ | ~ |
1-out-of-2 Delta-OT | malicious | KOS15 | 1.9* | ~ | ~ | ~ | ~ |
1-out-of-2 Delta-OT | passive | KOS15 | 1.7* | ~ | ~ | ~ | ~ |
1-out-of-2 | malicious | ALSZ15 | ~ | 17.3 | ~ | ~ | 10 |
1-out-of-2 | malicious | KOS15 | 3.9 / 0.7 | ~ | 1.1 | ~ | 2.9 |
1-out-of-2 | passive | IKNP03 | 3.7 / 0.6 | 11.3 | 0.6 | ~ | 2.7 |
1-out-of-2 Base | malicious | CO15 | 1,592/~ | ~ | ~ | ~ | ~ |
1-out-of-2 Base | malicious | NP00 | 12,876/~ | ~ | ~ | ~ | ~ |
The library is cross platform and has been tested on Windows, Mac and Linux. There is one mandatory dependency on Boost 1.69 (networking), and three optional dependencies on, Miracl, Relic or SimplestOT (Unix only) for Base OTs. Any or all of these dependenies can be enabled. See below.
In Powershell
, this will set up the project
git clone --recursive https://github.com/osu-crypto/libOTe.git
cd libOTe/cryptoTools/thirdparty/win
getBoost.ps1
cd ../../..
libOTe.sln
To see all the command line options, execute the program
frontend.exe
Boost and visual studio 2017: If boost does not build with visual studio 2017 follow these instructions.
Enabling Relic (for fast base OTs):
- Clone the Visual Studio port Relic
git clone --branch new_master https://github.com/ladnir/relic.git
. - Use the CMake command
cmake . -DMULTI=OPENMP -DCMAKE_INSTALL_PREFIX:PATH=C:\libs -DCMAKE_GENERATOR_PLATFORM=x64
generate a Visual Studio solution - Optional: Build with gmp/mpir for faster performance.
- Install it to
C:\libs
(build theINSTALL
VS project). - Edit the config file libOTe/cryptoTools/cryptoTools/Common/config.h to include
#define ENABLE_RELIC
.
Enabling Miracl (for base OTs):
cd libOTe/cryptoTools/thirdparty/win
getMiracl.ps1
(If the Miracl script fails to find visual studio 2017, manually open and build the Miracl solution.)cd ../../..
- Edit the config file libOTe/cryptoTools/cryptoTools/Common/config.h to include
#define ENABLE_MIRACL
.
IMPORTANT: By default, the build system needs the NASM compiler to be located
at C:\NASM\nasm.exe
. In the event that it isn't, there are two options, install it,
or enable the pure c++ implementation:
- Remove
cryptoTools/Crypto/asm/sha_win64.asm
from the cryptoTools Project. - Edit the config file libOTe/cryptoTools/cryptoTools/Common/config.h to remove
#define ENABLE_NASM
.
In short, this will build the project (without base OTs)
git clone --recursive https://github.com/osu-crypto/libOTe.git
cd libOTe/cryptoTools/thirdparty/linux
bash boost.get
cd ../../..
cmake .
make
This will build the minimum version of the library (wihtout base OTs). The libraries
will be placed in libOTe/lib
and the binary frontend_libOTe
will be placed in
libOTe/bin
To see all the command line options, execute the program
./bin/frontend_libOTe
Enable Base OTs using:
-
cmake . -DENABLE_MIRACL=ON
: Build the library with integration to the Miracl library. Requires building miraclcd libOTe/cryptoTools/thirdparty/linux; bash miracl.get
. -
cmake . -DENABLE_RELIC=ON
: Build the library with integration to the Relic library. Requires that relic is built withcmake . -DMULTI=OPENMP
and installed. -
Linux Only:
cmake . -DENABLE_SIMPLESTOT=ON
: Build the library with integration to the SimplestOT library implementing a base OT. Also works with only relic but is slower.
Other Options:
cmake . -DENABLE_CIRCUITS=ON
: Build the library with the circuit library enabled.cmake . -DENABLE_NASM=ON
: Build the library with the assembly base SHA1 implementation. Requires the NASM compiler.
Note: In the case that miracl or boost is already installed, the steps
cd libOTe/cryptoTools/thirdparty/linux; bash boost.get
can be skipped and CMake will attempt
to find them instead. Boost is found with the CMake findBoost package and miracl
is found with the find_library(miracl)
command.
You can either make install
on linux or link libOTe's source tree. In the latter
case, you will need to include the following:
- .../libOTe
- .../libOTe/cryptoTools
- .../libOTe/cryptoTools/thirdparty/linux/boost
- .../libOTe/cryptoTools/thirdparty/linux/miracl/miracl (if enabled)
- [Relic includes] (if enabled)
and link:
- .../libOTe/bin/liblibOTe.a
- .../libOTe/bin/libcryptoTools.a
- .../libOTe/bin/libSimplestOT.a (if enabled)
- .../libOTe/cryptoTools/thirdparty/linux/boost/stage/lib/libboost_system.a
- .../libOTe/cryptoTools/thirdparty/linux/boost/stage/lib/libboost_thread.a
- .../libOTe/cryptoTools/thirdparty/linux/miracl/miracl/source/libmiracl.a (if enabled)
- [Relic binar] (if enabled)
Note: On windows the linking paths follow a similar pattern.
Contact Peter Rindal rindalp@oregonstate.edu for any assistance on building or running the library.
Spread the word!
@misc{libOTe,
author = {Peter Rindal},
title = {{libOTe: an efficient, portable, and easy to use Oblivious Transfer Library}},
howpublished = {\url{https://github.com/osu-crypto/libOTe}},
}
The 1-out-of-N [OOS16] protocol currently is set to work forn N=276
but is capable of supporting arbitrary codes given the generator matrix in text
format. See ./libOTe/Tools/Bch511.txt
for an example.
The 1-out-of-N [KKRT16] for arbitrary N is also implemented and slightly faster than [OOS16]. However, [KKRT16] is in the semi-honest setting.
The approximate K-out-of-N OT [RR16] protocol is also implemented. This protocol allows for a rough bound on the value K with a very light weight cut and choose technique. It was introduced for a PSI protocol that builds on a Garbled Bloom Filter.
* Delta-OT does not use the RandomOracle or AES hash function.
** This timing was taken from the [OOS16] paper and their implementation used multiple threads. The number was not specified. When using the libOTe implementation with multiple threads, a timing of 2.6 seconds was obtained with the RandomOracle hash function.
It should be noted that the libOTe implementation uses the Boost ASIO library to perform more efficient asynchronous network IO. This involves using a background thread to help process network data. As such, this is not a completely fair comparison to the Apricot implementation but we don't expect it to have a large impact. It also appears that the Encrypto Group implementation uses asynchronous network IO.
The above timings were obtained with the follwoing options:
1-out-of-2 malicious:
- Apricot:
./ot.x -n 16777216 -p 0 -m a -l 100 & ./ot.x -p 1 -m a -n 16777216 -l 100
- Encrypto Group:
./ot.exe -r 0 -n 16777216 -o 1 & ./ot.exe -r 1 -n 16777216 -o 1
- emp-toolkit: 2x 223
./mot 0 1212 & ./mot 1 1212
1-out-of-2 semi-honest:
- Apricot:
./ot.x -n 16777216 -p 0 -m a -l 100 -pas & ./ot.x -p 1 -m a -n 16777216 -l 100 -pas
- Encrypto Group:
./ot.exe -r 0 -n 16777216 -o 0 & ./ot.exe -r 1 -n 16777216 -o 0
- emp-toolkit: 2 * 223
./shot 0 1212 & ./shot 1 1212
This project has been placed in the public domain. As such, you are unrestricted in how you use it, commercial or otherwise. However, no warranty of fitness is provided. If you found this project helpful, feel free to spread the word and cite us.
[IKNP03] - Yuval Ishai and Joe Kilian and Kobbi Nissim and Erez Petrank, Extending Oblivious Transfers Efficiently.
[KOS15] - Marcel Keller and Emmanuela Orsini and Peter Scholl, Actively Secure OT Extension with Optimal Overhead. eprint/2015/546
[OOS16] - Michele Orrù and Emmanuela Orsini and Peter Scholl, Actively Secure 1-out-of-N OT Extension with Application to Private Set Intersection. eprint/2016/933
[KKRT16] - Vladimir Kolesnikov and Ranjit Kumaresan and Mike Rosulek and Ni Trieu, Efficient Batched Oblivious PRF with Applications to Private Set Intersection. eprint/2016/799
[RR16] - Peter Rindal and Mike Rosulek, Improved Private Set Intersection against Malicious Adversaries. eprint/2016/746
[BLNNOOSS15] - Sai Sheshank Burra and Enrique Larraia and Jesper Buus Nielsen and Peter Sebastian Nordholt and Claudio Orlandi and Emmanuela Orsini and Peter Scholl and Nigel P. Smart, High Performance Multi-Party Computation for Binary Circuits Based on Oblivious Transfer. eprint/2015/472
[ALSZ15] - Gilad Asharov and Yehuda Lindell and Thomas Schneider and Michael Zohner, More Efficient Oblivious Transfer Extensions with Security for Malicious Adversaries. eprint/2015/061
[NP01] - Moni Naor, Benny Pinkas, Efficient Oblivious Transfer Protocols.