finextra, Pantor sets Fast protocol benchmarks, here. 2006 press release.
FIX Protocol, FPL Technical Roadmap for 2011, here.
Andersson Pantor: Slides seem competitive technically and tight to reality. For example, the latency map to 2011 system architecture seems plausible:
100 us 50 us 10 us 5 us 1 us …
100 us (one hundred microseconds) Tuned OS and application logic, optimized GbE network
50 us (fifty microseconds) Highly optimized application logic, real-time OS configuration
10 us (ten microseconds) Kernel bypass comms drivers, low memory footprint
5 us (five microseconds) 10G Ethernet, cache-aware application logic
1 us (one microsecond) Hardware assisted implementation?
and Andersson is presumably showing average test message sizes:
Test results FIX tag value Binary tag-value Simple FAST Full FAST
~200 bytes ~100 bytes ~35 bytes 15-25 bytes
4.6x compression w. FAST (not entirely sold on the compression idea architecturally, but you can compress and decompress in hardware reasonably fast I suspect)
< 100ns CPU overhead encode/decode per FAST formatted message (we were looking for ~60ns previously with Sandy Bridge)
Henry Young TS-Associates: Great diagram pg. 50 in the FPL Technical Roadmap; Buy Side, Broker, Venue schematic. See ts-a.com, here.