Soldb: In-Memory Database with SIMD-Accelerated Core Operations
Pugazhanti S
Dept. of Information Technology Sathyabama Institute
Chennai, India pugazhanti1@gmail.com
Naveen T K
Dept. of Information Technology Sathyabama Institute
Chennai, India naveen2005830@gmail.com
Mr.V.SARAVANAKUMAR
Dept. of Information Technology Sathyabama Institute
Chennai, India Saravanakumar.v.cse@sathyabama.ac.in
Abstract—Soldb is a specialized in-memory key-value database that achieves extreme performance optimization for four core operations—GET, SET, EXISTS, and DELETE—through aggressive application of ARM NEON SIMD vectorization, hardware-accelerated hashing, and cache- aware design on Apple Silicon (M1/M2/M3) architecture. Unlike general- purpose databases, Soldb adopts a focused approach: hyper-optimize funda- mental operations to their performance limits while maintaining durability through efficient copy-on-write snapshotting. The system leverages ARM CRC32C instructions for hash computation (0.5ns/byte), NEON intrinsics for parallel 16-byte key comparison, 128-byte cache-line aligned data structures, branchless execution paths, and lock-free atomic operations. Comprehensive benchmarking demonstrates sub-100ns P50 latency for GET operations (85ns), ultra-fast EXISTS checks using SIMD bloom filters (45ns), and throughput exceeding 11M ops/sec for reads and 8M ops/sec for writes on M1 Pro. Performance profiling confirms >95% L1 cache hit rates and <2% branch mispredictions. When compared against Redis 7.x for raw in-memory operation speed without networking overhead, Soldb achieves competitive performance (0.9-1.1x), with EXISTS operations showing 3.3x improvement through vectorized bloom filters. The snapshot mechanism uses fork()-based copy-on-write semantics with NEON-accelerated serial- ization, ensuring <1ms blocking time. This work demonstrates that extreme specialization and ARM-specific optimization can match production systems in core competencies, proving that focused optimization depth competes with breadth when scope is carefully constrained. The system serves applications requiring ultra-low latency key-value operations including caching layers, session stores, real-time analytics, and edge computing on ARM platforms.
Keywords—In-memory database, SIMD optimization, ARM NEON, Apple Silicon, key-value store, cache optimization, lock-free data struc- tures, high-performance computing
