Reamer Server is an institutional order-management core that you build your own trading system around, delivered as a library with a stable C interface. It holds order state, puts every strategy's orders into one strict sequence, runs your own pre-trade rules on every order and records every decision with its reason. On a 64-core AMD EPYC 9575F with release 4.6.0 it took 60 ns per order across its interface and peaked at 220,446 orders a second, with latency steady up to one strategy per CPU core. Your venues connect through connectors you own, in C, C++, Rust or Go, starting from the kit's worked FIX 4.4 integration.
It is a library you link into your own program. A short main() supplies your gate and your connector and starts it. That program is your server, shaped to your firm, without building or proving the core yourself.
The core, and the two parts you build
- The core (what you buy). It takes order intents from your strategies, puts them into one sequence, calls your gate once for each, and passes accepted orders to your connector. Orders can be market, limit, stop, stop-limit, trailing stop, trailing stop-limit, market-if-touched, limit-if-touched and market-to-limit orders, with day, GTC, GTD, IOC, FOK, at-the-open and at-the-close time in force. An order can also carry an attached take-profit and stop-loss, iceberg and minimum quantity, post-only, reduce-only, hidden and all-or-none flags, an OCO, OTO or OUO link to another order, an account and a destination, and up to 1,024 bytes of your own data. The core passes every field to your gate and connector, and your connector maps them to your broker. Fills and order updates go back to the strategy that sent the order, including fills on an attached exit.
- Your gate. Your pre-trade risk check. For each order it says pass or reject, with a reason, given the account's last known state. Its rules are yours; the core does not supply any. See How can I plug my own risk model into every order before it reaches the broker?
- Your connector. Your broker session, written for exactly the venue you trade. It sends orders, reports fills, and is the only source of positions and open orders. The kit's connectors run against a simulated venue and a paper broker; the connection to your real venue is the one you build.
Market data comes from your own feeds into your own strategy processes. The core handles orders.
Gate and connector can be written in C, C++, Rust or Go. Strategies connect over a socket protocol, specified byte for byte, from any language. See How do I run several strategies through one broker connection?
Performance and scale
- 60 ns per order across the C interface, and a median of 10.3 µs from a strategy's order to its accepted event, on a 64-core AMD EPYC 9575F with release 4.6.0.
- As many strategies as cores. Latency stays steady up to about one strategy per physical core, with about 200,000 to 220,000 orders a second in total on that one machine, tested from 1 to 512 strategies with every order matched to its event.
- Every decision on record. Each order is gated and recorded in one strict sequence, so what happened and why can be read back.
See How fast is it?
What else is in the kit
- Starting points. An accept-all gate with an in-memory paper broker, in C++ and Rust, and a worked integration in Go with a FIX 4.4 session, a strategy relay and a simulated venue that runs an order to a fill in one command.
- A live feed of every decision. Each acceptance, rejection (with your gate's reason), fill and order update is published on a shared-memory event stream that any number of your own processes can read. It holds a fixed number of recent events, so keeping a permanent log is up to your reader.
- Monitoring. A Prometheus
/metricsendpoint and a/healthroute. - A benchmark.
server-benchmeasures throughput and latency on your own hardware.
Scope
- Your broker, your connector. Any broker with an API, through a connector you control, with no vendor broker list to wait on.
- The broker is the record. On start, orders and positions are loaded from your connector. Anything else you want kept, you persist from the event stream.
- Mid-frequency order flow. The full order set above, from strategies on bars. Market data and the strategy itself come from your own processes.
- Linux x86-64, with glibc 2.39 or later.
How it compares
- QuantConnect, a hosted platform with data, backtesting and managed live trading through supported brokers.
- NautilusTrader, an open-source platform with venue adapters and its own order management.
- EPAM Deltix QuantOffice, an enterprise platform with over 100 exchange and broker connectors.
Price
$24,000 per seat per year, paid once, with no automatic renewal. The $2,000 30-day trial includes the full kit. See the product page, What is an order management engine? and pricing.