BrutalDB: An Embedded Object Database for .NET AoT Tick-Rate Workloads

This paper argues that the per-operation latency of a database access is determined first by where the database lives relative to the requesting process and only second by what the database engine does once the request arrives. For applications whose access pattern operates at tick-rate (game servers, edge sensor logging, financial intermediation, any program that reads and writes its own state thousands of times per logical frame), crossing a syscall, a loopback, or a network wall per operation imposes a latency floor structurally incompatible with the workload. The paper presents BrutalDB, an embedded object database for .NET applications targeting this workload class. BrutalDB uses ahead-of-time source generation to produce reflection-free hot paths, supports five index kinds against domain object types declared with attributes, and persists state via an append-only write-ahead log with optional AES-256-GCM encryption. The paper reports a per-operation latency comparison against PostgreSQL 18.3 and seven additional embedded configurations (SQLite in four durability modes including pure in-memory, LiteDB in two storage configurations) on a single x86_64 host with published reproducible benchmark harnesses. BrutalDB primary-key reads complete in fifty-six nanoseconds in memory mode, between one-and-a-half and six-hundred times faster than the comparator engines; the paper is also explicit about the two regimes where BrutalDB is not the right tool (cold-data point reads against disk-backed storage, and the current strict-per-commit-fsync implementation). The reference implementation has been in production use since 2021 as the persistence layer of a Custom MU Online MMORPG game server.