Case studies

Scaling a cross-DEX arbitrage desk

How a prop trading desk like “Apex Alpha” would move its arbitrage onto AllenHark infrastructure.

An illustrative scenario. The firm is a composite, and this page reports no measured results.

Two markets at two prices, and a stream looping between themAn abstract drawing of light on the page's dark ground. Two solid blocks stand apart, one tall and one short, two prices for the same thing; a bright stream arcs from the tall one to the short one and loops back.Two markets at two prices, and a stream looping between themAn abstract drawing of light on the page's dark ground. Two solid blocks stand apart, one tall and one short, two prices for the same thing; a bright stream arcs from the tall one to the short one and loops back.

The challenge

Apex Alpha has a working arbitrage strategy across Raydium, Orca and Meteora. Its bottleneck is the network: it reads the chain through public RPC and runs on general cloud hosting.

Too many of its transactions land after the price has moved, and a failed transaction still costs its fee.

The setup

Its own server

Execution moves to a single-tenant AMD server in Frankfurt, Amsterdam or Chicago. See the servers.

Pool changes as a stream

Account updates for the pools it trades arrive over Yellowstone gRPC instead of by polling.

A direct route to the leader

Signed transactions go to AllenHark Relay, which sends them to the current and upcoming leaders. The desk pays per transaction, with a tip inside each one, from 0.001 SOL.

What changes

The desk reads pool changes as they happen, decides on its own hardware and sends without the public route. We publish no results for this scenario; measure your own from your own servers.