New trading systems platform
Fri Jul 8 16:41:27 CEST 2005
On Fri, 8 Jul 2005, James Hague wrote:
> > What is wrong with using lists for this purpose?
> > You do process your time series sequentially, front to back, and I
> > think you can do it recursively as well. Erlang has tail recursion
> > and you can easily drop the first element of the list with [H|T].
> You were worrying about Erlang performance for general number
> crunching relative to C. Most of the time I'd say using lists is
> fine, but not if you're potentially talking about gigabytes of data
> and the kind of processing where a factor of 10 speed increase is very
Exactly!! Especially when it is time and $$ on the line,
those big shots on Wall Street won't move away from C/C++
even when they are ugly and hard to maintain...
> Erlang lists take twice as much space as tuples, and,
> much as I hate to admit it, if you just need lean and mean data
> processing on large and structurally simple data sets, then you'll
> always win in C (and by quite a lot, too).
In addition to financial world where data set is huge (but with
simple structures), another example is physics world where
an experiment at an accerlerator generates gigantic data sets.
The TTree class in ROOT (C++) package (from CERN) might
give an example as to how to organize (write with compression
and read with decompression) large data set.
> That said, I'd still try the raw Erlang solution and see how it goes.
For a small scale trading system, Erlang should be nicer
than the 'popular' TradeStation (with its ugly (macro) Easy Language).
More information about the erlang-questions