Investigation of Key-Value Model Performance in PostgreSQL and Redis: A Comparative Analysis of Read/Write Operations and Parallel Load Impact

An experimentally substantiated approach to comparing the performance of the key-value model implementation in PostgreSQL and Redis is proposed. Purpose: to evaluate the latency and throughput of read and write operations under varying degrees of parallel load. Methods: analysis of PostgreSQL architecture (MVCC, WAL, B-tree and GIN indexes, jsonb and hstore types) and Redis architecture (in-memory storage, strings and hashes, TTL, pipelining); creation of an experimental testbed; generation of a test table with 1,000,000 records; load testing using pgbench and redis-benchmark with parallelism from 1 to 128 clients; measurement of throughput, average and tail latencies (p95, p99). Results: Redis provides 5–10 times lower read latency and 3–5 times higher throughput compared to PostgreSQL for point GET/SET operations; the use of pipelining additionally increases throughput by 2–3 times; PostgreSQL retains advantages in versatility, supporting transactions, complex queries, and long-term storage. Practical significance: the developed testing methodology and obtained quantitative estimates are applicable when choosing a storage architecture for highload information systems, including monitoring and tracking systems in transport.