TTL Cache Benchmark (libttak)

A high-performance benchmark for libttak simulating a TTL-based cache with generational memory management.

Overview

This benchmark simulates a concurrent cache workload with:

• Generational Memory: Values are allocated in time-based epochs.
• Bulk Cleanup: When an epoch expires, all its memory is freed at once using mem_tree.
• Sharded Locking: Hash map is partitioned into shards for concurrency.
• Async Maintenance: Background task rotates epochs and cleans up old data.

Build

make

Run

./ttl_cache_bench [options]

Options

• --threads, -t: Number of worker threads (default: 4)
• --duration, -d: Benchmark duration in seconds (default: 10)
• --value-size, -v: Size of value blob in bytes (default: 256)
• --keyspace, -k: Number of unique keys (default: 100000)
• --ttl-ms, -l: TTL of items in milliseconds (default: 500)
• --epoch-ms, -e: Duration of one epoch in milliseconds (default: 250)
• --shards, -s: Number of map shards (default: 16)

Example

./ttl_cache_bench --threads 8 --duration 20 --value-size 1024 --shards 32

Metrics

• Ops/s: Total throughput (Get + Set + Del)
• HitRate: Percentage of GETs that found a valid item.
• AvgClean: Average time (microseconds) to destroy an expired epoch (bulk free).