Do I need more than one L-Micro Node to run multiple concurrent ML Trainings?

Based on the information provided in the extracts, it seems that QuantConnect's LEAN engine does not support concurrent training of multiple machine learning models within a single algorithm. The self.Train method is designed to run sequentially, not concurrently. This is likely due to the fact that running multiple concurrent training processes would require significant computational resources and could potentially lead to issues with memory management and performance.

In terms of running multiple algorithms concurrently, it appears that each algorithm requires its own server or node. This is mentioned in several of the forum discussions, such as this one where a user discusses running multiple live algorithms using paper trading. Each live algorithm requires its own live server or node, which comes at a cost.

If you're experiencing timeouts during backtesting when running concurrent ML trainings, it might be due to the computational intensity of training multiple ML models at the same time. As mentioned in this discussion, backtesting in QuantConnect is limited by the single thread speed, which could be a bottleneck when running multiple concurrent trainings.

In conclusion, if you want to run multiple concurrent ML trainings, you might need to consider running each training on a separate node or server. However, this would increase the cost. Alternatively, you could consider optimizing your ML training process to reduce the computational resources required, or explore other ways to achieve your goal within the constraints of the platform.

In that case, how to pass information and data between two or more nodes/servers?

Based on the information provided in the extracts, there isn't a built-in feature in QuantConnect's LEAN engine to directly communicate or pass data between two or more nodes/servers. As mentioned in this forum discussion, theoretically, you could create a control server that manages the communication between two algorithms, but this would likely require significant custom development and is not a feature provided out of the box by QuantConnect.

In another forum discussion, a user asked about allowing an algorithm to communicate with external software. The response from QuantConnect staff was that you can download signals/parameters from an external source to seed a universe or start your algorithm, but not upload signals/state. This suggests that while it's possible to pull data into an algorithm from an external source, pushing data out from an algorithm to an external source (or another algorithm) is not directly supported.

In conclusion, while it's theoretically possible to set up a system to pass data between different nodes/servers running different algorithms, doing so would likely require significant custom development and is not a feature directly supported by QuantConnect. You might want to consider other ways to achieve your goal within the constraints of the platform.

Can you please give more details about this ‘control server’? Like how to build it and how does it work?