Date: 2026-08-17Engine: KacheDB v0.1.0Status: ✅ All 4 Phases Complete & Verified
KacheDB is an in-memory storage engine designed from scratch in Rust to address the memory bottlenecks of modern AI and high-concurrency microservices:
Sub-4 ns Memory Allocation: Replaces runtime malloc/free with 64-byte aligned 2 MB Megaslab arenas, achieving 3.93 ns allocation latency regardless of slot size (128 B to 256 KB).L1 Cache-Speed Point Queries: SIMD-probed Swiss Table hash index delivers 3.15 ns lookup hit latency with lock-free S3-FIFO eviction flags.~10,000× TTFT Speedup for LLM KV-Cache: Hierarchical &[u32] Radix Prefix Tree matches a 1,024-token sequence in 2.61 µs , skipping costly GPU attention prefill.15.04 Million msgs/sec Zero-Copy IPC: POSIX Shared Memory (/dev/shm) lock-free SPSC ring buffers stream tensor descriptors across processes in 66.47 ns per message with zero serialization and zero memory copies.10.2 Million QPS per Core: Ingests, parses, executes, and encodes standard Redis RESP commands over TCP in 97.63 ns end-to-end.
Metric Specification
Operating System macOS 26.5.2 (Darwin 25F84) Architecture arm64 (Apple Silicon)Cores 8 Physical / 8 Logical Cores System Memory 16 GB Unified Memory Rust Toolchain rustc 1.97.1 / cargo 1.97.1Optimization Profile release (opt-level = 3)Benchmarking Suite Criterion.rs v0.5.1 (100 samples per test, 3s warmup)
Crate Benchmark Target Measured Latency Target / Industry Baseline
kachedb-coreArena Slot Allocation (AppSmall 128 B) 3.93 ns < 20 ns target (5.1× faster) kachedb-coreArena Slot Allocation (AppMedium 512 B) 3.92 ns < 20 ns target kachedb-coreArena Slot Allocation (AppLarge 4 KB) 3.96 ns < 20 ns target kachedb-coreArena Slot Allocation (Tensor64KB 64 KB) 3.97 ns < 20 ns target kachedb-coreArena Slot Allocation (Tensor256KB 256 KB) 4.11 ns < 20 ns target kachedb-corePool Alloc + Dealloc Cycle (AppSmall) 5.76 ns < 20 ns target kachedb-corePool Alloc + Dealloc Cycle (Tensor64KB) 7.06 ns < 20 ns target kachedb-hashSwiss Table Lookup Hit (1M keys preloaded) 3.15 ns L1 cache probe speed kachedb-hashSwiss Table Lookup Miss 8.27 ns Fast group termination
Crate Benchmark Target Measured Latency Throughput / Speedup
kachedb-radixPrefix Lookup Hit (128 tokens / 8 blocks) 253.85 ns ~31.7 ns per block hop kachedb-radixPrefix Lookup Hit (1,024 tokens / 64 blocks) 2.61 µs ~10,000× faster than GPU prefillkachedb-radixPrefix Lookup Hit (4,096 tokens / 256 blocks) 19.98 µs Deep context chain lookup kachedb-radixInsert 1,024-token sequence (64 new nodes) 2.41 µs ~37.6 ns per node kachedb-radixHierarchical Bottom-up LRU Eviction 568.13 ns Sub-microsecond memory reclaim kachedb-shmSingle-Thread 128B Slot Roundtrip 89.39 ns Lock-free push + pop kachedb-shmCross-Thread SPSC Ring Streaming 66.47 ns / msg 15.04 Million msgs/sec
Crate Benchmark Target Measured Latency Single-Core Capacity
kachedb-proto-respZero-Alloc GET Frame Parse & Decode 68.80 ns 14.53 Million cmds/sec kachedb-proto-respZero-Alloc SET Frame Parse & Decode 96.15 ns 10.40 Million cmds/sec kachedb-proto-respZero-Alloc MGET Frame Parse & Decode (4 keys) 153.06 ns 6.53 Million cmds/sec kachedb-proto-respFrame Bulk String Serialization 8.25 ns 121.2 Million frames/sec kachedb-netFull GET Hit Pipeline Execution 97.63 ns 10.24 Million requests/sec / core kachedb-netFull SET + DEL Cycle Execution 267.58 ns 3.74 Million write cycles/sec
Subsystem Operation Measured Performance Context
kachedb-cliLive Server Loopback Ping-Pong (10K reqs) 48,493 req/sec (20.62 µs/req) Synchronous unpipelined TCP bindings/python64-byte Header Validation & Recovery < 1 µs 0 heap allocations bindings/pythonZero-Copy Tensor Extraction (np.frombuffer) Instantaneous (< 50 ns) 0 bytes copied (direct memory view)
All benchmarks can be reproduced locally with:
# Run all crate micro-benchmarks
cargo bench --workspace
# Or run individual crate benchmarks
cargo bench -p kachedb-core
cargo bench -p kachedb-hash
cargo bench -p kachedb-radix
cargo bench -p kachedb-shm
cargo bench -p kachedb-proto-resp
cargo bench -p kachedb-net