Cache benchmarks
The v12 benchmark suite focuses on the typed codec that backs Redis cache entries. The runtime has three injected cache managers—memory, Frozen snapshots and Redis—but cache entries are never represented by shared global state.
Codec hit and miss
Benchmark class: CacheCodecBenchmarks
| Benchmark | What it measures |
|---|---|
EncodeDto | Deterministic DTO binary encoding |
CodecCacheHit | Decode of a previously encoded typed payload |
CodecCacheMissAndPopulate | Miss, encode and population work without an external Redis round trip |
The generated codec is AOT-safe and carries a codec identity and contract hash. Redis adds a versioned envelope around this payload. An incompatible or corrupt envelope is a controlled miss, removed best-effort, and counted separately; it is never deserialized as an older DTO contract.
What is intentionally outside this microbenchmark
An IConnectionMultiplexer is reused at application scope, while IDatabase is lightweight. Real Redis latency, retry behavior and server throughput are covered by the opt-in redis category rather than this CPU-only codec result. It needs both CAERIUS_BENCHMARK_SQL_CONNECTION and CAERIUS_BENCHMARK_REDIS_CONNECTION, uses a disposable database/Redis endpoint, and provisions only dbo.caerius_benchmark_redis_reader. It is excluded from all and the portable gate; no values are logged.
Likewise, Frozen cache reads are snapshot lookups: a miss does not build a new dictionary. Memory cache population uses local single-flight work and expiration policy. Benchmark these operations in a host that reflects the application's expected cache size and concurrency.
Run the codec suite with:
dotnet run -c Release --project benchmarks/CaeriusNet.Benchmark.csproj -- --filter "*CacheCodecBenchmarks*"