HackMDLow Latency C++ for Fun and Profit
重點
- remove branch
- template-based configuration
- memory alloc (pooling)
- exception
- almost zero-cost if not throw
- avoid multi threading
- shared data
- synchronization
- contention (last level cache lock)
- SPSC lockfree queue
- cache line friendly
- dense_hash_map / flat_hash_map / ska::flat_hash_set
- gcc
- branch prediction hints
- force inline / noinline
- attribute hot & cold
- prefetching
- hotpath pre warm (run through the entire system)
- performance counter
- cache miss event etc
- compare instruction
- RDTSC for high resolution timestamp
