Light Speed Trading

[Joy Pathak]

Logic Gates: The Next Phase Of High-Frequency Trading

I thought companies were using logic gates already? Maybe not for this specific type of application I guess.

 

[agency]

Wow, that article is dripping with technological illiteracy. Computers are arrays of logic gates, they're the fundamental building blocks.

I'm guessing they're trying to talk about the use of FPGAs or ASICs in high-frequency trading but they should really pass their articles by someone with a cursory knowledge of computers before they publish stuff like this. Logic gates! Cutting edge stuff.

 

[Joy Pathak]

Wow, that article is dripping with technological illiteracy. Computers are arrays of logic gates, they're the fundamental building blocks.

I'm guessing they're trying to talk about the use of FPGAs or ASICs in high-frequency trading but they should really pass their articles by someone with a cursory knowledge of computers before they publish stuff like this. Logic gates! Cutting edge stuff.

Well I was wondering more about the usage of single chip to run one process comment. I thought companies had dedicated systems to run single computations already.

 

[agency]

Yeah, I definitely think the application specific hardware is an interesting (and inevitable) development in a place where performance is so critical, definitely an interesting area of work. This article made it seem like logic gates were somehow cutting edge technology.

 

[Stefan Zota]

Well I was wondering more about the usage of single chip to run one process comment. I thought companies had dedicated systems to run single computations already.

I agree with agency the article seems to be written about a guy that read somewhere 0 AND 1 = 0

The discussion around overall execution speed is broad, but the first item to consider: where is the bottleneck? CPU is just one component, in most cases in front-end trading execution systems it's not the main criteria. Things much more important are thread context switching/memory access etc ...

 

[Noko]

I see two major problems with this:

1. The author fails to realize that the typical hierarchy of a program is that a program requires coreutils (operating system more or less) which require a kernel which requires the hardware itself. To increase speed, the author suggests that the entire algorithms be moved to chips similar to the CPU. A better way would be to utilize what we already have; clusters of Xeon servers and GPU processing which are incredibly fast. How do we utilize that in a more efficient manner? Write a new kernel. The fastest method we can achieve right now would be to, in ASM, write exactly what we need, no more, no less. If your algorithms can't be translated to ASM, then a light weight C/C++/Fortran compiler can be written, and by light weight I mean it leaves out that which you don't need. See problem two on why this is better.

2. When you write a program, do you let it run forever, or do you have to make occasional changes? "Oh, something went wrong, let me just go move around some logic gates on the nanoscopic scale."

Looking back, is this a joke article? Did he also write about aluminum-powered cars being released next year just in case it happened?