Why does Node not support several event loops in one Process?

Question

As the title says, why does Node not support several event loops in one Process? The idea, is that node spawns a user decided number of threads, which each have an event loop. When a new callback is added, it is scheduled to one of the threads loops. The reason, probably against this idea is because now each thread can cause mutation and other bad stuff, but you could make the user decide what threads share state and so have to be run on the same thread or have a message like event memory sharing concept. This would help with computationally heavy tasks, this would reduce the overhead of sharing data between node instances to achieve the same result and therefore achieve better performance for the hardware and more user requests.

To make it more clear, when I mean several event loops, then that would be like forking and creating several node instances to utilise multiple cores but instead node itself would spawn 4 threads all with one event loop, so there is a lower cost of communication because it is using threads instead if instances if you are doing lot's of data sharing, not workers that you spawn to do heavy work and have nothing to do with the event loop.

Answer 1

As soon as you have several threads running in the same process and sharing the same virtual address space, you got synchronization issues and you'll probably need to have mutexes (e.g. to deal with the case of several thread loops accessing the same object concurrently).

^{(I am assuming a Linux operating system with pthreads below)}

So you'll need to extend the Javascript language for that; either you'll transparently add mutexes into every object making them heavier (a pthread_mutext_t has 40 bytes on my Linux/Debian/x86-64) and slower (you'll need to call something like pthread_mutex_lock on every access), or you put that burden on the user, forcing him to add additional code, and adding a new way of undefined behavior, and possibility of deadlock.

At last, multi-threaded garbage collection is quite difficult. Read the GC handbook for more.

BTW, on many Linux systems, processes are nearly as efficient and easy and fast to start as threads (see clone(2) used by pthreads and sometimes as substitute to fork(2)...), so having several dozens of Node.js processes makes sense. Also, local IPC (e.g. thru pipe(7)-s or unix(7)-s sockets) is very efficient on Linux, so the communication costs can be quite low (usually more than ten times faster than Ethernet on a local network).