Beyond Sums and Differences: Homomorphic Encryption Gets a Comparative Edge

Author: Denis Avetisyan

A new technique streamlines privacy-preserving computation by efficiently combining arithmetic and comparison operations within fully homomorphic encryption schemes.

This paper introduces ‘space switching’ to significantly improve the performance of BGV/BFV-based schemes for secure computation over Zp/Zpr.

Existing fully homomorphic encryption (FHE) schemes struggle to efficiently support both arithmetic and comparison operations within a unified framework, hindering practical privacy-preserving computation. This paper, ‘Efficient Arithmetic-and-Comparison Homomorphic Encryption with Space Switching’, introduces a novel ‘space switching’ technique to seamlessly integrate these operations within $\text{FV}-style$ schemes by dynamically transitioning between number spaces $\mathbb{Z}_{p^r}$ and digit spaces $\mathbb{Z}_{p}$ . Our approach achieves up to $17\times$ faster performance than scheme switching and $15\times$ faster than direct comparison on database workloads, demonstrating a significant step toward practical FHE deployment. Could this space switching technique unlock new possibilities for complex, privacy-preserving data analysis across diverse application domains?

The Persistent Challenge of Homomorphic Operations

Fully Homomorphic Encryption $FHE$ offers the compelling vision of performing computations on encrypted data, safeguarding privacy without requiring decryption. However, realizing this potential faces a persistent challenge: the efficient combination of different types of operations. While $FHE$ schemes excel at certain arithmetic – addition and multiplication – crucial comparison operations, such as determining if one encrypted value is greater than another, often present a significant performance bottleneck. This disparity stems from the fundamentally different mathematical properties required for each operation; arithmetic benefits from relatively straightforward polynomial evaluations, while comparisons necessitate more complex techniques like bootstrapping or approximations. Consequently, hybrid operations – those requiring both arithmetic and comparisons – frequently suffer from substantial overhead, limiting the practicality of $FHE$ in real-world applications that demand both data processing and decision-making capabilities.

Traditional Fully Homomorphic Encryption (FHE) schemes encounter substantial challenges when tasked with performing both arithmetic and comparison operations on encrypted data. These operations possess fundamentally different requirements; arithmetic benefits from precision and efficient polynomial evaluation, while comparisons necessitate bitwise operations and often involve bootstrapping-a computationally expensive process of refreshing ciphertexts. This disparity creates performance bottlenecks, as optimizing for one type of operation frequently degrades the efficiency of the other. Consequently, many practical applications requiring a blend of these operations-such as private database queries involving both numerical calculations and conditional filtering-become prohibitively slow or resource-intensive with existing FHE implementations, thereby limiting the broader applicability of this promising privacy-enhancing technology.

Current methods for enabling both arithmetic and comparison operations within Fully Homomorphic Encryption (FHE) frequently necessitate trade-offs between computational efficiency and accuracy. Many implementations depend on approximating complex functions with simpler ones to reduce processing demands, which can introduce quantifiable errors into the final results. Alternatively, some approaches involve switching between different FHE schemes – each optimized for a specific operation type – but this introduces significant overhead due to the need for repeated encryption and decryption, and the complexities of ensuring data consistency across schemes. These approximations and scheme switches, while attempting to address performance limitations, ultimately compromise the precision and reliability of computations performed on encrypted data, hindering the widespread adoption of FHE in practical applications requiring verifiable results.

A Streamlined Approach: Dynamic Ciphertext Spaces

The Space Switching method achieves unified arithmetic and comparison operations by operating within two distinct plaintext spaces: $Z_p$ for comparisons and $Z_{pr}$ for arithmetic, where p is a prime number and r is an integer. Data is dynamically transitioned between these spaces as needed; comparisons are performed directly on values represented in $Z_p$ , while arithmetic operations are conducted on values within $Z_{pr}$ . This dynamic switching eliminates the necessity for either complex scheme alterations or approximate computations, allowing for a streamlined process that avoids the overhead typically associated with handling different operational requirements simultaneously.

Traditional secure multiparty computation often necessitates switching between cryptographic schemes or employing approximations to manage the computational load associated with different operations. Space Switching eliminates this requirement by performing both arithmetic and comparisons within dynamically adjusted plaintext spaces – $Z_p$ and $Z_{pr}$ – without necessitating these complex transitions. This simplification directly reduces computational overhead as it avoids the performance penalties associated with scheme changes and the potential inaccuracies introduced by approximations. By operating natively within these spaces, the method streamlines the overall computation process, contributing to improved efficiency and predictable results.

The Space Switching method employs a reduction step to convert data between the $Z_p$ and $Z_{pr}$ spaces, and a subsequent modulus-raising step to maintain computational correctness. This conversion process utilizes digit extraction techniques to efficiently represent data in the target space. Specifically, the digit extraction allows for the isolation of relevant coefficients necessary for the reduction and modulus-raising operations, minimizing the number of computations required during the space transition. The design of these steps ensures that arithmetic operations are performed in $Z_{pr}$ while comparisons occur in $Z_p$ , without requiring data expansion or approximation.

Empirical Validation: Performance Gains with TPC-H

The TPC-H benchmark is a decision support benchmark consisting of a suite of business analytic queries that stress various aspects of a database management system. It utilizes a volume-based metric, scaling data set size from 1GB to 10TB, and evaluates performance across a range of ad-hoc queries involving selections, joins, aggregations, and sorts. Its widespread adoption within the database research community allows for standardized and comparable performance evaluations, providing a robust methodology for assessing the efficiency of data processing techniques like Space Switching. The benchmark’s queries are designed to mimic real-world business intelligence tasks, making it a practical and relevant measure of database system performance.

Performance validation using the TPC-H benchmark demonstrates that Space Switching provides a significant improvement over traditional database methodologies. Comparative analysis reveals a maximum performance increase of 17x on standard database workloads when utilizing Space Switching, as opposed to direct comparison techniques. This improvement is observed across various query types and data volumes representative of typical data warehousing applications, indicating a substantial reduction in query response times and increased throughput.

Performance validation using the TPC-H benchmark demonstrates that our method achieves a 10x speedup on standard database workloads when compared to scheme switching. Beyond typical database operations, a significant performance increase was observed in LT (lookup table) operations; specifically, a 136x speedup was measured for LT operations utilizing 20-bit input compared to direct comparison methods. These results indicate substantial gains in both general database performance and specialized lookup table functionality.

Utilizing Space Switching, LT operations performed on 8-bit data achieved an amortized runtime of 4.34 milliseconds. Amortized runtime calculates the average time per operation over a series of operations, accounting for any initial setup costs distributed across subsequent executions. This measurement indicates that, on average, each 8-bit LT operation completes in 4.34ms when leveraging the Space Switching methodology, representing a performance benchmark for this specific data size and operation type.

Performance validation using the TPC-H benchmark demonstrated a significant speedup with Space Switching on LT operations when processing datasets containing 2¹⁴ rows. Specifically, results indicate a performance improvement ranging from 10x to 20x compared to scheme switching for these operations. This speedup was consistently observed across multiple test runs and represents a substantial gain in processing efficiency for larger datasets, highlighting the benefits of Space Switching in managing and accelerating database workloads at scale.

Toward Scalable Privacy: A Vision for the Future

The Space Switching technique represents a significant advancement in the pursuit of practical Fully Homomorphic Encryption (FHE). Traditionally, FHE schemes suffered from substantial performance bottlenecks due to the growth of ciphertext size during computations. This method addresses this limitation by strategically switching between different ciphertext spaces, effectively controlling the noise that accumulates with each operation. By carefully managing these transitions, Space Switching minimizes ciphertext expansion, allowing for more efficient and scalable encrypted computations. This foundational approach not only improves the performance of existing FHE schemes but also paves the way for the development of novel cryptographic constructions capable of handling increasingly complex data processing tasks without compromising security – a crucial step toward widespread adoption of privacy-preserving computation.

Ongoing research prioritizes streamlining the computationally intensive reduction and modulus-raising operations inherent in Fully Homomorphic Encryption (FHE) schemes. These steps, critical for maintaining ciphertext integrity and managing noise growth, currently represent significant performance bottlenecks. Investigations are underway to explore novel algorithmic approaches and hardware acceleration techniques tailored to these specific calculations. Simultaneously, efforts are directed towards broadening the applicability of these optimizations by adapting them for use with diverse FHE schemes beyond the initial implementation, fostering a more unified and efficient landscape for secure computation. This expansion aims to overcome the limitations of scheme-specific optimizations and unlock the full potential of FHE across a wider range of cryptographic applications and data types.

Significant performance gains in Fully Homomorphic Encryption (FHE) are anticipated through the incorporation of advanced techniques like Functional Bootstrapping and the strategic use of Look-Up Tables. Functional Bootstrapping allows for the efficient reduction of ciphertext noise without decrypting the underlying data, dramatically accelerating computations and reducing communication overhead. Simultaneously, the implementation of Look-Up Tables enables the rapid evaluation of complex functions directly on encrypted data, bypassing computationally expensive operations. By pre-computing and storing the results of these functions in an encrypted form, FHE systems can perform intricate calculations – such as those involving polynomial evaluations or neural network inferences – with substantially improved efficiency and scalability, ultimately broadening the scope of practical applications for privacy-preserving computation.

The pursuit of efficiency, as demonstrated by this work on space switching within fully homomorphic encryption, echoes a fundamental principle. It seeks not elaborate complexity, but streamlined functionality. The paper’s integration of arithmetic and comparison operations, enhancing performance in privacy-preserving computation, exemplifies this. As Paul Erdős observed, “A mathematician knows a lot of things, but the mathematician who knows the most knows the least.” This echoes the sentiment-removing unnecessary layers to reveal the core truth of computation. The elegance lies in doing more with less, a reduction toward clarity, particularly within the intricate landscape of schemes like BGV/BFV.

What Lies Ahead?

The introduction of ‘space switching’ offers a necessary reduction in the baroque complexity that has long plagued fully homomorphic encryption. The field has, for too long, chased performance through accretion – layering optimization upon optimization. This work suggests a more fruitful path lies in fundamental reassessment of how arithmetic and comparison are integrated, not merely accelerated. The gains achieved are, however, not a destination, but a clearer view of remaining obstacles.

The current reliance on schemes like BGV/BFV, while providing a solid foundation, inherently limits scalability. The question is not simply how to make these schemes faster, but whether a fundamentally different approach – one that minimizes the overhead of ciphertext manipulation – is required. Exploring alternatives to polynomial-based constructions, or perhaps a more disciplined application of modular arithmetic beyond Zp/Zpr, warrants consideration. The elegance of a solution will likely reside in its parsimony.

Ultimately, the true measure of progress will not be benchmarks, but usability. A fully homomorphic encryption scheme, however efficient, remains an academic exercise if it cannot be readily deployed in practical applications. The focus must shift from demonstrating what is possible to delivering what is useful-a subtle, but critical, distinction. Further research should prioritize simplification, standardization, and the development of tools that abstract away the underlying complexity, allowing practitioners to focus on the data, not the cryptography.

Original article: https://arxiv.org/pdf/2604.19890.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

The Persistent Challenge of Homomorphic Operations

A Streamlined Approach: Dynamic Ciphertext Spaces

Empirical Validation: Performance Gains with TPC-H

Toward Scalable Privacy: A Vision for the Future

What Lies Ahead?

See also: