[mangrish]

Cool! Very interesting.

I think given the languages you have selected you are coming more from a quant background? These languages are great for heuristics and analysis but you would really want all 'static' components such as connectivity built in assembly/C/C++. For 'algo' components I like Java as you can still pull microsecond order latencies when crunching numbers but more importantly it gives you a huge time to market advantage than the C/C++ for almost the same speeds. I'm also not clear on if you are connecting directly to the market for market data or using aggregation (like reuters). The latter would be too slow. I'm also not clear on what middleware you are using which is probably the biggest decision you will have to make. Most either use inhouse tech or 29west LBM (everyone still calls it that even though they were bought out).

An overlooked part of HFT in my experience OS optimisation and even things like TCP bypass (for some components) which can lead to huge speed advantages and end to end latency reductions. I agree with those about FPGA.. in my experience they really don't come into the equation except for components that rarely change.

Anyway a few guys including Martin Thompson have felt similarly to you and initiated the lodestone project (http://lodestonefoundation.wordpress.com/). If you are keen to learn more their architectures for low latency, distributed and componentisation then I feel that you could join forces and contribute to that initiative. FYI: Martin built most of the technology behind the disruptor (http://lmax-exchange.github.io/disruptor/).

Great to see interest in making this knowledge more widely available :)

[zygomega]

thanks mangrish,

quant. How could you tell? I'm conecting with an aggregator cause the direct market feeds are monopolistic price gougers. It's an easy switch if we make it up that curve.

I'm not sure if I even understand what middleware is (I'm a quant), but I think the answer is the disruptor!

[mangrish]

From the languages.. I've worked with lots of quants and they all rave about R!

Yeah..the aggregator is where you get done in both cost and latency. The prop shops and hedge funds pay through the nose for that stuff so unless you come packing a little capital, true HFT is an issue.

On the middleware, not really.. so you would have a market data component that will be pushing stuff to various components (real time risk, the pricer, and the trading engine). The disruptor sits on the 'in' queue to those components.. the middleware is what pushes messages between instances running on the same/different machines.

Hope that helps :)

[zygomega]

I get the sense that the Lodestone project is now defunct.