Costs of Encrypted Computation: Why Fully homomorphic computations are Slow
Over the years, data privacy rules in several geographies have evolved and several countries are strictly mandating organizations to ensure data privacy regulation and enhance transparency on storage and processing of customers’ data. Non-compliance to these regulations is causing huge financial and credibility implications to organizations. Currently, 10% of data is covered under privacy regulations – this is expected to be 65% by 2023. Also, with organizations increasingly relying on cloud service providers (CSPs), the major challenge for CSPs is to protect privacy and confidentiality of the data while still being able to cater to users’ needs.
Fully homomorphic encryption (FHE), an evolving approach with mathematically provable security guarantees, enables computations on the encrypted data; thus, offering protection to the privacy of data. As privacy regulations are critical for both organizations as well as CSPs, FHE enables both of them to reduce their liabilities. While FHE solves the problem with adequate security guarantees, it incurs some cost in terms of computational complexity and memory requirement. As remarked in our previous white paper on this subject, FHE based computation is a million times slower than normal computation on plaintext. We now intuitively describe several fundamental aspects that hinder performance and how the computing model in FHE differs from the normal scenario. This can help explain the limitations in a simpler manner, which in turn can help users envision new applications in FHE domain.