# Expected Performance
With this architecture:
- 50,000 - 200,000+ msgs/sec (depending on message size)
- Sub-millisecond latency for push/pop operations
- O(1) ack/nack/renew latency regardless of queue depth (DLV-0014)
- O(n_queues) queue stats — never O(total_messages), on either storage backend (SYS-0018)
- Full durability with RocksDB's WAL
- Bounded memory usage proportional to hot tier size, not total queue depth
- Payloads stored in mmap'd files — OS page cache manages hot/cold eviction
- jemalloc allocator reduces fragmentation under high-churn workloads
- Linear scalability with queue count
- Adaptive memory management with container-aware pressure detection
- Graduated pressure response: Warning (85%) shrinks block cache; Critical (90%) sheds publishes
- Hysteresis prevents thrashing at pressure boundary (enter 85%, exit 75%)
- Dynamic RocksDB block cache resizing under pressure with automatic recovery