Micro and transport layer security benchmarking of post-quantum versus classical cryptography

Abstract

As quantum computing advances, widely deployed public-key cryptosystems face a credible long-term risk, motivating a transition to post-quantum cryptography (PQC). Building on NIST’s standardization, this paper evaluates the practical performance of CRYSTALS-Kyber and CRYSTALS-Dilithium through both microbenchmarks and TLS 1.3 deployments, with the expectation that compute costs remain competitive while deployment overhead is dominated by larger handshake messages. At the algorithm level, we measure key generation, encapsulation/decapsulation, signing, and verification for reference and AVX2-optimized implementations using a consistent toolchain and runtime configuration, and compare them against classical baselines (RSA, ECDH, and ECDSA). Across tested parameter sets, AVX2 delivers multifold speedups and keeps operations in the sub-millisecond range, with Kyber and Dilithium execution times that are competitive with or better than security-aligned classical counterparts. At the protocol level, we integrate ML-KEM and ML-DSA into TLS 1.3 and extract handshake latency (ClientHello→ServerHello and ClientHello→END) and handshake message sizes from captured traces in a low-RTT setting. PQC handshakes incur only a modest latency increase at the 128-bit level relative to P-256, while at 192- and 256-bit security they complete substantially faster than P-384 and P-521 configurations. The main cost is bandwidth: server-to-client handshake traffic increases from a few kilobytes in classical TLS to roughly 8–15 kB under PQC. Overall, the results indicate that deployment challenges are driven less by computation than by larger handshake footprints, especially on bandwidth- or RTT-constrained links.