low latency - How fast is state of the art HFT trading systems today? -

Whenever you hear about high frequency trading (HFT) and how fast algorithms are I thinking - these What is the day fast?

Update

I am not thinking of deviation which is due to physical distance between an exchange and the server is running a business application But latency started by the program

Being more specific: In an application on that application, the timing of the event on the wires indicates the order / price on the wire? To wit. Tick-to-Trade Time

Are we talking sub-dials? Or sub microseconds?

How do people get these latencies? Coding in the assembly? FPGAs? Good old C ++ code?

Update

An interesting article has recently been published on ACM, which provides a lot of detail in today's HFT technology, which An excellent reading:

You have received very good answers. , Though - the most al-Qaeda secret is that you just do not know how fast it is in both ways - some people can not tell you how fast they work Are, because they do not want. Other, let's say, "exaggerated" for many reasons (attracting investors or customers for one).

Rumors about picocytes, for example, are rather insulting, 10 nanoseconds and 0.1 nanoseconds are exactly the same, because when it takes time to reach the trading server it is much more than that.

And, most importantly, though you have not asked, if you go about trying to do business with algorithms, then do not try to be fast, try to be smart Have seen algorithms that can handle full seconds of latency and too many can earn money.