Low latency trading is not trivial but it is not rocket science either. Latency goes up in two ways: 1. Physics - you are further away from the exchange (each mile of cable adds 5 microseconds of latency) and 2. Technology (how fast is your hardware and software in processing signals).

Many providers focus on one or the other and make you work hard on top of it. One common example is when technology is fast but the algo is physically far separated form the exchange. The separation could be via distance or number of hops / servers. If your vendor has a feed handler or a risk control reside on a separate machine, congratulations - they add separation at no extra charge (you pay with extra 10 mics instead). If your strategy is executed on your own machine that communicates directly with the exchange, then your firm will absorb all monthly costs (co-loction space, marked data fees, system management etc.) plus you either would have your own skilled (expensive) developers or pay asinine fees to a software vendor on top of it all.

Is there a better way?

We asked ourselves what NOT to do to make DMA fastest and accessible to more traders.

First one is simple, the algorythms must run on the same machine that connects to the exchange, we do not add extra separation - physics is physics.

We build our software with a few things in mind: 1. Data transmission, 2.Processing and 3.Presentation.

We use ithe best available technology and coding methods to ensure that:
1. Operating system (kernel) is not involved in any of those steps (OS is slow as it spreads attention on a million things).
2. There is no copying of data anywhere on the data flow path (copying adds unnecessary work).
3. There is minimal buffering of incoming data (if a packet waiting in queue it does not feel great, think of the last time you stood in a line).

That's it, its common sense and we use it.