Memory usage skyrocketed with nim 0.18.0 (in my async tcp service test)

[aguspiza2]

I have done a little test to find out if I could run some microservices with 
async code in memory constrained devices. Here is the test code:

[https://gist.github.com/aguspiza/80e34b5cf65aa3bbfd19c7339ee9b695](https://gist.github.com/aguspiza/80e34b5cf65aa3bbfd19c7339ee9b695)

That test uses 5x more memory with 0.18.0 than with 0.17.2 (or any previous 
version at least with boehm gc)

Here you have some numbers:
    
    
    Linux x64 0.17.2 boehm 24MB used (release 14MB)
    Linux x64 0.18.0 boehm 94MB used (release 84MB)
    
    other gc's have similar increases +40MB:
    0.17.2 default gc 70MB (release 30-36MB)
    0.18.0 default gc 146MB (release 110MB)
    similar result with gc:v2 146MB (release 110MB)
    

I am getting those numbers from RSS column on "ps"
    
    
    $ ps aux
    USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
    root     13946 38.9  0.0  18024 15136 pts/0    Sl+  11:55   0:22 ./httpasync
    

Is this a known issue?

Is this expected from latest gc optimization (TLSF algorithm) in 0.18.0?

Why is there so much difference between gc's?

Should I use boehm, markandsweep, v2 or default gc?

Boehm is giving the best results but Araq told me (at IRC) that it is not using 
Nim's memory allocator ?!?!