ADAM BRAZDA

Systems Engineering SIU '26

Divalia

Experimental_Cryptographic_Framework

Successor to Noiseblind // Substitution-Permutation Network

Execution_Environment

C++20 // AVX2 SIMD // AES-NI // SHA-NI // Argon2 // CLI11 // CMake

Investigation_Report

Divalia emerged as the successor to Noiseblind, abandoning simple obfuscation for a rigorous Substitution-Permutation Network. The architecture commits fully to a "hardware-first" philosophy—no scalar fallbacks, pure SIMD execution.

Performance profiling exposed an unexpected ceiling: the operating system itself. With CPU execution ports saturated, kernel page fault handling became the dominant bottleneck. This validated the aggressive micro-optimizations—the software was no longer the limiting factor.

The integrity subsystem posed a unique challenge: maintaining deterministic MAC generation across wildly different concurrency models. The solution binds threads to pre-allocated Merkle lanes, ensuring cryptographic consistency whether processing via mmap work-stealing or barrier-synchronized I/O.

⚠ Experimental framework. Novel cryptographic constructions without formal audit. Academic use only.

Optimization_Log

// Hardware_Saturation

Pushed the Angerona Kernel to ~350MB/s sustained throughput. Profiling revealed OS Page Faults consuming 34% of runtime—the system now waits on the kernel, not the CPU.

// Instruction_Pipelining

Strategically ordered SIMD operations to exploit instruction-level parallelism. Pair-wise element processing and deliberate register scheduling minimize pipeline stalls and maximize execution unit utilization.

// Deterministic_Integrity

Engineered a 32-lane Merkle Tree for parallel MAC computation. Each thread owns dedicated lanes, guaranteeing bit-identical checksums regardless of thread count or I/O strategy.

Source_Code_Archive