alex.ks
|
|
| Total Posts: 1 |
| Joined: Aug 2009 |
| |
|
Me and my friends consider designing a low latency (target 10-20 usec from wire to memory) FPGA-based board for high frequency trading that can be directly plugged into a co-located server in the PCI express slot. The product will include the hardware with several network interfaces, firmware with 2-3 feed handlers and order routers (e.g. Nasdaq ITCH/OUCH, Arca Direct/ArcaBook, NYSE Ultra), and API to control the FPGA from a software solution.
If you have experience in high frequency trading your opinion would be highly appreciated.
(1) Do you think such a product is needed?
(2) Would a hedge fund or a brokerage company buy or test this product?
(3) What is a realistic price for this product?
Thanks
Alex |
|
|
|
 |
FDAXHunter
|
| Founding Member |
| Total Posts: 8114 |
| Joined: Mar 2004 |
| |
|
You're not the first person to think of this. This comes up routinely and I've seen more than one person experiment with specialized hardware in order to get a speed advantage in some way, shape or form.
Have you heard of HyperFeed (now merged with exegy)? These guys have been pushing dedicated hardware for years, without that much success.
The problem is, that actually, it's not only speed that counts, but the ability to rapidly adapt your software to your own ideas. The ability to do things slightly differently (and better) is where a large part of your edge in electronic trading lies. This notion is obviously very much orthogonal to anything embedded in hardware where you have no control and no access.
So my take on your questions is:
1) No, but I'm sure there's someone that says yes.
2) Not really, though the old adage: "A sucker born every minute" does apply. Undoubtedly you can find someone that thinks this is just super-cool and will shell out 100,000 USD for it without blinking. You can probably even find 10 suckers like that in finance. If you build it, you will find some buyers.
3) You're not going to be selling very many of these, so you'd have to charge a very large item price to make it worth your while.
|
Salman Pushdie |
|
 |
nikol
|
|
| Total Posts: 311 |
| Joined: Jun 2005 |
| |
| |
|
|
Yeah, what FDAX said.
Even more to the point, you rapidly approach a point of diminishing returns when designing latency out of trading infrastructure. Network latency is easily of the order of tens of microseconds, too, even in colos.
Carefully look at a "day in the life" of a trading decision, end to end, and you'll likely find many other places where driving out latency is more cost effective. |
|
|
|
svquant
|
|
| Total Posts: 107 |
| Joined: Apr 2007 |
| |
|
Just some random tidbits...
* I do know a high frequency shop in Chicago did buy a custom FPGA solution for speeding up their data feed in order to improve trading. Person who told me this was a bit guarded on the details but this company either used it to increase capacity within their current system (perhaps not to buy more co-location space?) or to shave time off for their algs.
* There was a company out in San Jose that was building a specialized computer/accelerator for finance applications in order to sell to Wall St firms. After some initial success with beta testing the company & its investors decided to go into the high frequency trading business using the hardware itself. Figured out why make $500K a system selling to Wall St vs making $500K a day trading :-) Well that was the story... No idea how they are doing and if this business model change worked out.
* Need to keep an eye on internal efforts since there are now many good tools that can turn C-code into an FPGA. In fact Microsoft Research has mentioned how they have used one of these tools in order to speed up some inner loops of websearch for something like a 100x or greater speed up (in this part of the code) vs OO C++ for less than 2 months investment including learning time on the new software...
So I see firms/people rolling their own if there is any true advantage to hardware speed up in ones trading flow.
|
|
|
|
|
