Squeezing more coroutines ?

[Bogdan Harjoc <harjoc@...>]

Hi,

While writing a performance testing tool, I chose Lua to describe a set 
of tests that
run in parallel.

So far things are looking great, as far as both available cpu and memory 
usage
are concerned: after increasing LUAI_MAXCSTACK, I've been testing with 
300.000
concurrent and mostly idle threads, created with lua_newthread(). Memory 
usage
for these is about 350M, i.e. a little over 1KB per created thread.

After a glance at the lua_State definition, it looks like there a couple 
of members
that could be left out but nothing that would noticeably reduce 
sizeof(lua_State).

Has anyone managed to create something on the order of 1,000,000 
coroutines ?
If so, any tips on what to leave out from lua_State (or other places) 
would be
appreciated.


Cheers!
Bogdan