You could download the C++ version of QuickFix and look at their examples. This is the "real deal" -- the type of programming that many brokerage firms, exchanges, and HFT's use (although not with the freeware QuickFix version of FIX). However, it's also true that the time commitment to become proficient at this is significant compared to other solutions such as Ninja or Tradelink. Java is another one that's similar to C++ but a little easier (with no memory management), and also has many broker-supplied API's, not to mention FIX implementations (such as the free QuickFixJ). I disagree with the posters who say you can't learn C++ as your first language. If you're already making some headway, there's no reason you can't continue. It's all just a matter of what you're looking to accomplish. Give a serious look at the alternatives and make your own decision.
My next question is about market data. Any recommendations on a broker / data provider with an API and good historical data? Perhaps one that has a paper trading account that will allow me to track the programs success? No point in writing a program if I can not test it. What does everyone use?
If your goal is to be rich, then take a shortcut. I handed a one pager to an exhibitor and in 15 minutes he had a system running and the Sharpe he calculated was over 60. He did the work twice since he didn't believe the results on his first pass. Use any drag and drop logic system. Blox is what he was selling. Lets say you are dealing with more leveraged approaches than 4 to 6 day cycles of stocks that only return 1000% a year on capital. (See dark horse's calculation of one beginner using the rule set the Blox guy used.) Here you would be trading intraday. To do the logic for such a system, you can warm up by reading an English translation of a Polish book on Haskell. It is humorous and you will get it down in no time. Then you switch to another language called SQL. I typed my tables for SQL in two days (22 tables, initially). SQL books abound. The point of creating Tthe APL languages in the early 70's was to support the principles of how markets operate (there are five). The reason you went to C stuff was because of the environment you work or learn in. You can check out the range of annual returns and they are what they are. You may want to get rich slowly which is the tradition of your environment, apparently. The invention of RDBMS dosn't work in that tradition. The ball breaker you are facing is whether to learn how markets work and then learn how to use the best support system. Almost no one would agree with me. They are data processing as you think you must. Try to understand what is going on for only 15 minutes. Just think for a very brief time and see if you can "get it". On only one sheet of paper, draw all the different R's of RDBMS for just two bars. There are but 10. 8 of them reduce to the other 2. See if you can throw away the sheet of paper and go back to your present environment. If you can't dispose of it, at least write the movie script this weekend. The transition you might be making is more rivetting than Bourne. Write it down.
Most people never see the markets. Most vendors will not export data to you from their visualizing platforms. So you go directly to the markets to get the data supply. So you have two sources: one to look at and one to feed your coded ATS. as I said the 1000% a year that took 15 minutes of dragging and dropping just used EOD data and the "builder" used the NAZ 100 since he could not form a selective Universe for testing. The fixed costs of trading are Trivial. So are the variable costs of participating. Because market variables are always "coarse", so is the data. This granularity dictates that logic is the mathematics of the markets. See the one page test you did over 15 minutes. If you want to get up to speed rapidly, just go to a site that posts delayed data and have a ball. The bar to rise above in your planned work is the zigzag chart on any fractal. Fractals are interlocking and the ratio, then is fixed. It is 3:1 and you should be able to use intermediate form (grade 5 to 7) geometry to get the interlocking ratio of 3 to 1.
I would try an easier language before committing to learn C++. You'll have to learn alot , from memory management to object oriented coding to debugging. I've been using www.stockTesting.com to backtest. You can write a strategy in JavaScript , backtest & auto trade all from a web browser
Your criteria match IB : interactivebrokers. As to language: everything is C++ in one form or another. It is good idea to know the basic concepts of it but it is professional coder language build for coder productivity. For fishing expedition it is better to use platform independent Java or .net C#. It is not only language but coding environment and API your code will work with.
Excel, VBA, DDE, Interactive Brokers. Minimum tech effort. Maximum "getting started" productivity. If you wear that out, only then bother with the serious engineering type stuff.