[P1] Benchmarks vs RocksDB (throughput, latency, memory)

[varun29ankuS]

Context

To credibly position shodh-redb as a RocksDB alternative, we need head-to-head benchmarks on standard workloads. Without numbers, adoption is based on faith.

Scope

• Benchmark workloads:
  • Sequential write: 1M, 10M, 100M keys
  • Random write: same key counts
  • Sequential read: point lookups after bulk load
  • Random read: point lookups with uniform distribution
  • Range scan: scan 1K, 10K, 100K contiguous keys
  • Mixed read/write: 80/20, 50/50, 20/80 ratios
  • Large values: 1KB, 10KB, 100KB value sizes
• Metrics per workload:
  • Throughput (ops/sec)
  • Latency (p50, p95, p99, p999)
  • Memory usage (RSS)
  • Disk usage after compaction
  • Write amplification
• Compare against RocksDB via rust-rocksdb bindings
• Run on standardized hardware (document specs)

Acceptance Criteria

• Benchmark harness using criterion or custom framework
• All workloads implemented for both shodh-redb and RocksDB
• Results published with hardware specs
• Identified workloads where shodh-redb wins and where it loses
• Performance regression CI (track key metrics over time)