•Systems engineer•1 min read
AeroLog
A zero-latency telemetry engine for HFT-style systems — lock-free shared-memory IPC, asynchronous sidecar logging, and nanosecond-scale hot-path writes.
C++LinuxPOSIX Shared MemoryPythonCMake

AeroLog started from a simple systems question:
What if logging never touched the trading thread at all?
Traditional logging pipelines are full of hidden latency — mutex contention, syscalls, kernel scheduling, disk waits. For most software, that's fine. For latency-sensitive systems, those microseconds add up fast.
So instead of optimizing file writes, I separated the logging pipeline into two worlds:
- the Hot Path, where the producer only writes to shared memory
- the Cold Path, where a separate sidecar process handles persistence asynchronously
The producer never waits for disk. It just moves forward.
The architecture
At the center of AeroLog is a lock-free Single Producer Single Consumer ring buffer built inside POSIX shared memory.
struct SharedBuffer {
alignas(64) std::atomic<uint64_t> head;
alignas(64) std::atomic<uint64_t> tail;
std::atomic<bool> producer_done;
LogEntry entries[RING_SIZE];
};