Generating secure random numbers with openssl
October 24, 2014
I started working on a new R package with bindings for OpenSSL. The initial release is now available from CRAN. To install the package on Linux you need
libssl-dev (Debian/Ubuntu) or
openssl-devel (Fedora, RHEL, CentOS). For Mac and Windows, precompiled binaries are available from CRAN as usual. The Mac version is compiled against the version of OpenSSL that is included with OSX. See the comments in Makevars if you want to compile against a more recent version of OpenSSL.
Secure random numbers
The initial release of openssl implements bindings to the OpenSSL random number generator, which will be used to generate session keys in the upcoming version of the OpenCPU system. This feature was requested by Ruben Arslan who noted that the default RNG in R is not suitable for this because it is predictable and lack of entropy can lead to collisions. I’m not a crypto expert but it seems like everyone uses OpenSSL for secure RNG, hence this new package. For implementation details, see the respective OpenSSL documentation pages.
rand_pseudo_bytes functions return a raw vector with random bytes:
To convert them to integers (0-255) simply use
Or convert bits to booleans:
Mapping random bytes to a continuous distribution requires a bit of math. For example to combine four 8bit bytes into a single 32bit double from the standard uniform distribution:
And from U(0,1) we can map into draws from a probability distribution using its CDF:
However note the native R random number generators are much faster and have better numeric properties. Also the OpenSSL RNG is not intended for generating large sequences of random numbers as often used in statistics. It is mainly useful in situations where it is critical to create a little bit of secure randomness that can not be manipulated. Typical applications include encryption keys, drinking games, or raffle drawings at your local R user group.
More fun stuff
OpenSSL has a lot of other useful stuff which we coud add to the R package in future versions. In particular public key methods to sign and verify packages is something that R and CRAN could really benefit from. Simon Urbanek is working on something similar as well in the PKI package, which also builds on OpenSSL.
If you you would like to see some other OpenSSL functionality in the R package, feel free to send a pull request with bindings on github. It would be great to have people involved with better understanding cryptographic methods.