Work
Multi-venue execution
One execution engine connected to 5 venues, with risk limits on every order. Built for clients and run in-house.
Result
<20 µsinternal tick-to-order across 5 venues
Our in-house system executes across 5 venues at internal tick-to-order under 20 µs.
Investment firms, plus our own in-house system
Price gaps across venues close in microseconds, and risk limits must hold on every order.
In production
exec-gateway · sessions2026-09-22 14:26 UTC
| venue | protocol | session | seq out | orders / s | limit |
|---|---|---|---|---|---|
| V1 | FIX 4.4 | logged on | 2,418,907 | 142 | 500 |
| V2 | SBE binary | logged on | 9,113,254 | 388 | 1,000 |
| V3 | OUCH 5.0 | logged on | 655,031 | 61 | 400 |
| V4 | WebSocket | connected | 1,207,448 | 97 | 300 |
| V5 | FIX 4.4 | logged on | 418,622 | 23 | 200 |
pre-trade risk: position, notional and rate limits on every order · kill switch armed
exec-core · hot path breakdownrolling 60 min · HW timestamps
feed decode SBE / ITCH, zero-copy
1.32 µs
book update per-venue L2
0.71 µs
signal strategy callback
2.86 µs
risk check position, notional, rate
0.18 µs
order encode FIX / SBE / OUCH
0.38 µs
wire tx kernel bypass NIC
1.25 µs
| stage | p50 µs | p99 µs |
|---|---|---|
| feed decode | 1.32 | 2.40 |
| book update | 0.71 | 1.50 |
| signal | 2.86 | 6.10 |
| risk check | 0.18 | 0.41 |
| order encode | 0.38 | 0.80 |
| wire tx | 1.25 | 2.20 |
| tick to order | 6.70 | 12.30 |
end-to-end p99 12.3 µs against a 20 µs budget · stage p99s do not add up to the total
risk-gateway · pre-trade limits2026-09-22 14:26 UTC
| venue | orders / s | rate limit | notional (USD m) | limit | used |
|---|---|---|---|---|---|
| V1 | 142 | 500 | 4.18 | 10 | 42% |
| V2 | 388 | 1,000 | 7.92 | 20 | 40% |
| V3 | 61 | 400 | 1.35 | 5 | 27% |
| V4 | 97 | 300 | 2.61 | 8 | 33% |
| V5 | 23 | 200 | 0.44 | 3 | 15% |
Orders checked today2,904,118every order, before the wire
Rejected1612 price band · 3 notional · 1 position
Check time0.41 µsp99, inside tick-to-order
Kill switcharmedper venue and global
Representative views of production tooling. Data anonymized.
What we built
- Every venue in one market view
- C++ and Rust hot path, tick to order
- Risk check on every order, one session per venue
C++RustExchange and broker APIsGrafana
Next case
Next step
Have a hard system to build?
Tell us what it has to do. We will tell you how we would build it.