> That still relies on IPC, and the associated kernel overhead to set, ctl and access the shared memory, IN ADDITION to the kernel overhead of having to do a co

> That still relies on IPC, and the associated kernel overhead to set, ctl and access the shared memory, IN ADDITION to the kernel overhead of having to do a context switch between processes. In multithreading there is no need for a context switch. Also memory is automatically shared by all threads. The arena model can be completely managed by Python. I don't get why the SC has not chosen this. Well, I do get it, but it don't like what it implies.