CRYPTOCOMP

State of the art

The current trend for data outsourcing, with the commercial success of services like DropBox or Google Drive (or more generally all the social networks) is easily understandable: letting someone else host your data frees you from many technical issues, and this for a nominal price. However, with this trend come many new security challenges:

• How does one guarantee the confidentiality of the data with respect to other users?
• And confidentiality with respect to the service provider?
• What about data integrity or user privacy?

These problems existed before outsourcing became popular, but this particular context intensifies them even more.

The simplest solution to ensure the confidentiality of outsourced data is to encrypt it. This is very easy to implement as long as outsourced data is a simple backup with no real purpose beyond protecting against data-loss. But the strength of data outsourcing comes from the additional services offered by the service providers: indexing, sharing, collaborative work… This requires the service to perform computation on your outsourced data, which is impossible if it is encrypted, unless Fully Homomorphic Encryption is used!

Fully Homomorphic Encryption allows to perform arbitrary computations on encrypted data, in a sense solving the problem of secure cloud computing. Unfortunately, existing Fully Homomorphic Encryption schemes are fairly inefficient, with only asymptotic performance analysis. It is thus difficult to assess how far we are from their practical implementation in security solutions.

Objectives of the project

The first goal of the project is to select the best FHE solutions that exist and try to improve their performance as much as possible. This task is closely related with the task of properly selecting parameters and sizes that match an expected security level.

Then, the main goal of CRYPTOCOMP is to develop a fully automated “source to source” compiler allowing to convert any function into a “secure” version of this function able to run on encrypted data. The main features of this compiler should be:

• complete transparency of the underlying fully homomorphic encryption scheme used
• definition of low-level computation models (possibly down to levels such as Turing machines or random access machines) suitable for the efficient execution of certain algorithm classes (sorting, signal processing, etc.)
• analysis and regulation of data dependent flow control to allow the service provider access to a given part of the data, and only to this part of the data
• automatic and efficient parallelization (through the compiler) of the cryptographic computations

Expected outcome

By the end of the project, we expect to have a fully functional demonstration platform available. Users will be able to upload their encrypted private data, locally compile their normal program into a fully homomorphic program and outsource the execution of this program to the demonstration platform. Once the execution is complete, the user can download the encrypted result from the platform and decrypt it.

Related presentations & talks

Related publications