Quote from danzman:
The main pitfalls would be time and cost. Instead of rebuilding the wheel, it's often a great idea to use an existing package as a foundation, and then use DLL functions to implement whatever you're trying to do (I'm assuming that you can't do what you're trying to do with Tradestation or Ninjatrader or the many others available).
Now if you're trying to do some type of cloud computing...the homemade route is probably better than buying 1000 licenses.
D
That level of strategy develop you describe if far easier with TZ simply because the platform is open source. Plus, TZ is committed to the best performance possible. The licenses already allow multiple CPUs. If someone really has a way to make cloud computing speed the processing, then we'll adjust the licensing to allow that.
But the reality, of testing trading is that there are serious limitations to how much work you can do in parrallel.
That's because any serious historical test on multiple symbols or strategies will want to have portfolio level position sizing and risk management.
In that case every strategy instance that's running depends on all the others to adjust position sizing.
What really slows down parrallel processing is the intercommunication and synchronization between threads.
If that is done as a "cloud" on multiple separate servers then the commication between them will be extremely high volume and across a slow network.
It's far more efficient to use multiple threads on a multi-core CPU.
TickZoom is looking into using a GPU but it's unclear if that will be a performance gain.
The real frontier in optimizing is still in speeding up/optimizing the software rather than throwing more hardware at the problem.
Sincerely,
Wayne