QuantConnect Community

In my experience, you want to use quantconnect's free data as opposed to IB's data unless you are already subscribed and pay for some specialized data from IB that quantconnect doesn't have. The ram is the amount of server memory that your algorithm can take up before it errors out. Things that use up your memory are extensive history calls or using a lot of securities simultaniously. If you are trading a small group of symbols intraday then you should have no memory problems. All the algorithms have the same CPU power allocated to them so it's not like you can get faster execution by buying more. The only way to ensure your execution is as fast as possible is to write well optimized code. Remember that latency doesn't matter until you get to the realm of HFT which is measured in milliseconds or less, in which case Quantconnect probably isn't what you're looking for.

The algorithms are designed to be continuously deployed but they stop executing after market hours and resume the next morning automatically as no new data is coming in. I belive that scheduled functions can run any time you designate. If your system liquidates EOD then you want to schedule that function so that it leaves enough time (a few minutes) to execute your trades and ensure you get fills before the market closes. 

The algorithm will occasionally get errors during live trading so you should make it stateless so that you can redeploy it when needed and it will pick back up where it left off.

Thank you Lukel. I think my algo is not "stateless" in the sense that it extensively relies on some dataframes I am caching in the Initialize method. If the algo gets errors and tries to resume, will it try to reinitialize? Let me know if I am totally missing the concept of "stateless"

Thank you for the informative discussion!
Let me just add the link to the documentation on Live Trading before we close this thread: