I wrote a FIX Engine that can process order execution with a round-trip time of 5.5µs.

So to answer your question, that means the processing time client -> server including parsing, serialising and network code is only RTT/2 = 2.25µs

Given parsing and serialisation can be massively optimised thanks to SIMD, most of the time will be spent in the network code. This is where Linux kernel tuning and/or TCP Kernel bypass technologies (OpenOnLoad) really help.

more details on www.fixisoft.com