Fork me on GitHub

What is the typical number of virtual peers on a host? For containerized deployment, would a container run a single virtual peer?


Just found the following in the API documentation: > In a production environment, you should start by booting up N virtual peers for N cores on the physical machine.


@joshg: Yes, roughly number of peers per machine == number of cores. A container should have all the peers inside it, not just 1.


If you have tasks that are rarely used (eg. flow conditions tasks that take error values), then these will be mostly idle, so the peer per core is a rough estimate


I'll improve the docs there a little. I think we could mention that typically this is because it's one main task lifecycle processing thread per task, and that stateful tasks and input tasks often require more


Thanks. Virtual peers are described as processes, but I wasn't sure if that meant process in the abstract sense or actual OS processes.


Yeah, really they're a few threads, which we should do a better job of describing. Mostly of the work is done by one thread


That makes sense. Thanks for the clarification.