You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
Fastest Technology for HF Auto Trading
- Thread starter Loki45
- Start date
The topic is not really worth pursuing unless you have $$$ for a staff of data pro's to process, interpret, test and execute ... which means $100k's for each employee...
Software like KDB+ to start this process in a HF setting is $100k+ and you need a super expensive data feed...
Good luck..
Software like KDB+ to start this process in a HF setting is $100k+ and you need a super expensive data feed...
Good luck..
http://www.solarflare.com/index.php
http://www.fusionio.com/
then get guys to program for the hardware itself.
http://www.fusionio.com/
then get guys to program for the hardware itself.
http://www.xilinx.com/products/devices.htmQuote from rosy2:
http://www.solarflare.com/index.php
http://www.fusionio.com/
then get guys to program for the hardware itself.
About two years ago I heard about some guys that looked at implementing high frequency algorithms for a small hedge fund using tools from http://www.mitrionics.com/. Not certain if they did target Xilinx FPGA though.
Haven't heard about them since. Maybe they're still in the lab...
Unfortunately... my info is second hand... (I'm guessing rosy's is in 1st... considering his competency from the past)... (Also... the topic at hand is completely out of my league... the most I personally do is GPU prog. in C++ or C-compile MKL...)Quote from Hugin:
About two years ago I heard about some guys that looked at implementing high frequency algorithms for a small hedge fund using tools from http://www.mitrionics.com/. Not certain if they did target Xilinx FPGA though.
Haven't heard about them since. Maybe they're still in the lab...
That said...
What I was told is that a FPGA does provide better performance but is not exactly cost effective compared a cluster of UNIX/C-Compiled servers, solely based on the lack of engineers with experience and the time it takes to get the FPGA programmed to be optimal to the surrounding architecture. (I guess testing and profiling it is a pain in the arse...)
The FASTEST Technology is, from what I've heard is, CPLD/FPGA added with GPU.
Quote from TSGannGalt:
Unfortunately... my info is second hand... (I'm guessing rosy's is in 1st... considering his competency from the past)... (Also... the topic at hand is completely out of my league... the most I personally do is GPU prog. in C++ or C-compile MKL...)
That said...
What I was told is that a FPGA does provide better performance but is not exactly cost effective compared a cluster of UNIX/C-Compiled servers, solely based on the lack of engineers with experience and the time it takes to get the FPGA programmed to be optimal to the surrounding architecture. (I guess testing and profiling it is a pain in the arse...)
The FASTEST Technology is, from what I've heard is, CPLD/FPGA added with GPU.
I'm no FPGA expert but from what I've heard the solution Mitrionics provides is to convert code written in a variant of C into something that can be loaded into the FPGA. This should make it easier both to find developers and to verify the code provided that the platform includes some test tools.
Even so it is probably a lot of work but if performance is key I guess this could compete with a GPU based solution, depending on the structure of the trading algorithm. My guess is only a handful of people/firms would benefit from this type of solution.
I've never dealt with programmable chips stuff and ran tests so I can't say much... and honestly... I wouldn't touch personally... I'd have something else handle them... But it would be helpful to know how they are.Quote from Hugin:
I'm no FPGA expert but from what I've heard the solution Mitrionics provides is to convert code written in a variant of C into something that can be loaded into the FPGA. This should make it easier both to find developers and to verify the code provided that the platform includes some test tools.
Even so it is probably a lot of work but if performance is key I guess this could compete with a GPU based solution, depending on the structure of the trading algorithm. My guess is only a handful of people/firms would benefit from this type of solution.
Hopefully our resident IT masters (nitro and rosy...) can provide some insights regarding this...
Define High frequency?
Do you want to process and place orders in microseconds, milliseconds or seconds, minutes etc?
With Standard PC hardware, connecting to your broker's back office and price feeds over the internet you can build a reliable trading system with redundancy and precision to 200 ms.
By precision I mean the ability to reliably process price feeds, place orders, confirm fills and manage fail overs.
If you need precision inside of 200 ms you should look at co-locating at the exchange/broker data center. You should be able to run as fast as they can.
For < 3 ms to microsecond precision you need some experts and serious hardware. Your trading application should interface directly with the exchange and clearing house. Try contacting Advantage Futures, they have the infrastructure to support these types of efforts.
Do you want to process and place orders in microseconds, milliseconds or seconds, minutes etc?
With Standard PC hardware, connecting to your broker's back office and price feeds over the internet you can build a reliable trading system with redundancy and precision to 200 ms.
By precision I mean the ability to reliably process price feeds, place orders, confirm fills and manage fail overs.
If you need precision inside of 200 ms you should look at co-locating at the exchange/broker data center. You should be able to run as fast as they can.
For < 3 ms to microsecond precision you need some experts and serious hardware. Your trading application should interface directly with the exchange and clearing house. Try contacting Advantage Futures, they have the infrastructure to support these types of efforts.