Hey rg,
I cant say a lot about Fix 5.0 or FAST 1.2; however, I believe those are both network protocols. Smoke is about sharing data among multiple threads running on the same processor. The Change Control Manager idea could be extended to work over a network, but this would introduce increased latency. The concurrency is fast and reliable (we did analysis for deadlocks, race conditions, etc). However, Smoke was designed for games, not mission critical applications. You can still apply the concepts to your field of expertise, but please do your own validation. The AI system is easy enough to rewrite for more traditional AI (e.g. for data stock analysis). This framework was designed for parallelism on full function and powerful cores (e.g. an x86 core) and parallelism scales well with the number of cores. Parallelism for something like CUDA would require major rewrites of code (especially in regards to data management and propagation). I hope this information helps. Please keep me informed about your work with Smoke :D
Thanks,
Orion



ultra hi-frequency stock market trading?
hi
anybody use this kind of thing for ultra hi-frequency stock market trading? might sound silly, i don't know...???? is the messaging faster that FIX 5.0 or FAST 1.2 protocols? Is the AI somehow more integrated or enhanced for statistical learning than compteing technologies??
how fast and reliable is the concurrency? vs. erlang level robustness? vs CUDA or Cilk++ speed? how many porcessors are needed before the distribution cost is outweighed by the speed advantage? More than 4, or in the hundreds?
I am truly interersted and able to deploy if there is an advantage here -
rg