exchange-core

Ultra-fast market exchange core matching engine based on LMAX Disruptor and Eclipse Collections (ex. Goldman Sachs GS Collections).

Capable to process 5M order book operations per second on 8-years old hardware (Intel® Xeon® X5690) with moderate latency degradation:

rate	50.0%	90.0%	95.0%	99.0%	99.9%	99.99%	worst
125K	0.6µs	0.9µs	1.0µs	1.4µs	4µs	24µs	41µs
250K	0.6µs	0.9µs	1.0µs	1.4µs	9µs	27µs	41µs
500K	0.6µs	0.9µs	1.0µs	1.6µs	14µs	29µs	42µs
1M	0.5µs	0.9µs	1.2µs	4µs	22µs	31µs	45µs
2M	0.5µs	1.2µs	3.9µs	10µs	30µs	39µs	60µs
3M	0.7µs	3.6µs	6.2µs	15µs	36µs	45µs	60µs
4M	1.0µs	6.0µs	9µs	25µs	45µs	55µs	70µs
5M	1.5µs	9.5µs	16µs	42µs	150µs	170µs	190µs
6M	5µs	30µs	45µs	300µs	500µs	520µs	540µs
7M	60µs	1.3ms	1.5ms	1.8ms	1.9ms	1.9ms	1.9ms

Benchmark configuration:

Single symbol order book.
3,000,000 inbound messages are distributed as follows: 9% GTC orders, 3% IOC orders, 6% cancel commands, 82% move commands. About 6% of all messages are triggering one or more trades.
1,000 active user accounts.
In average ~1,000 limit orders are active, placed in ~750 different price slots.
Latency results are only for risk processing and orders matching. Other stuff like network interface latency, IPC, journalling is not included.
Test data is not bursty, meaning constant interval between commands (0.2~8µs depending on target throughput).
BBO prices are not changing significantly throughout the test. No avalanche orders.
No coordinated omission effect for latency benchmark. Any processing delay affects measurements for next following messages.
GC is triggered prior/after running every benchmark cycle (3,000,000 messages).
RHEL 7.5, network-latency tuned profile, dual X5690 6 cores 3.47GHz, one socket isolated and tickless, spectre/meltdown protection disabled.
Java version 8u192, newer Java 8 versions can have a performance bug

HFT optimized. Priority is a limit-order-move operation mean latency (currently ~0.5µs). Cancel operation takes ~0.7µs, placing new order ~1.0µs;
In-memory working state.
Lock-free and contention-free orders matching and risk control algorithms.
No floating-point operations, no accuracy loss.
Matching engine and risk control operations are atomic and deterministic.
Pipelined multi-core processing (based on LMAX Disruptor): each CPU core is responsible for certain processing stage, user accounts shard, or symbol order books shard.
Low GC pressure, objects pooling.
Two different risk processing modes (per symbol): direct-exchange and margin-trade.
Supports crossing Ask-Bid orders for market makers.
Two implementations of matching engine: reference (naive) and performance-optimized.
Testing - unit-tests, integration tests, stress tests, integrity tests.
Automatic threads affinity (requires JNA).
State snapshot (serialization) and reconstruct operations.

greatyoungshaw / exchange-core Goto Github PK