4 ms·

I see C is missing. If you're going to do a language comparison, might as well include the one language we know will have the best performance and lowest memory

gigel82 3y ago
There is no light-weight threading concept in standard C/C++, and the cost of one OS thread is at least 1Mb (stack) + 4Kb (tls) reserved memory (at least by default, on modern Windows), so for 1m threads that will be 1Tb of RAM :)
- shrimp_emoji 3y ago
  How is your compiler addressing bits like this?
- tialaramex 3y ago
  C++ 20 has async/await (but of course named co_async and co_await, because C++ isn't the sort of language where you're allowed nice things, the beatings will continue until morale improves) However AIUI C++ 20 doesn't actually supply an executor out of the box, so you would need to choose what to do here as for Rust, where they picked both tokio and async-std and you can see they have different performance. C does only have threads, but you could presumably pull off the same trick in C that Rust does, to get Linux to give you the bare minimum actual resident memory for your thread and you needn't care about the notional "real" size of the thread since we're 64-bit and address space is basically free.
  - foderking 3y ago
    > (but of course named co_async and co_await, because C++ isn't the sort of language where you're allowed nice things, the beatings will continue until morale improves) LMAO
    - gpderetta 3y ago
      co_<whatever> has become a meme in the C++ community.
- pkolaczk 3y ago
  Stack memory is allocated lazily on Linux. I thought this benchmarks shows it pretty clearly. Otherwise 10k threads would be already in GBs territory, when in fact it is below 100 MB.
kaba0 3y ago
Since when is C the most performant? One can’t even write a performant generic vector with it. C++ is the de facto language used for hight performance.
- remram 3y ago
  "One can't even write a performant garbage-collected function in C++" You're describing features not performance.
  - kaba0 3y ago
    So how do you write programs without data structures? You copy paste some implementation and modify it slightly at every use site? Possible optimizations depend on the semantics of the language — C++ can do just orders of magnitude more with for example a string data structure, than what is possible with a C “string”. So no, C is not the most performant language in general — de facto C++ is used for everything performance critical.
    - remram 3y ago
      > You copy paste some implementation and modify it slightly at every use site? Yes, or use macros. Not saying it's the most convenient, but that's an orthogonal concern. You wouldn't say hand-written assembly isn't performant, so your objection strikes me as wrong.

gigel82 3y ago

There is no light-weight threading concept in standard C/C++, and the cost of one OS thread is at least 1Mb (stack) + 4Kb (tls) reserved memory (at least by default, on modern Windows), so for 1m threads that will be 1Tb of RAM :)

shrimp_emoji 3y ago
How is your compiler addressing bits like this?
tialaramex 3y ago
C++ 20 has async/await (but of course named co_async and co_await, because C++ isn't the sort of language where you're allowed nice things, the beatings will continue until morale improves) However AIUI C++ 20 doesn't actually supply an executor out of the box, so you would need to choose what to do here as for Rust, where they picked both tokio and async-std and you can see they have different performance. C does only have threads, but you could presumably pull off the same trick in C that Rust does, to get Linux to give you the bare minimum actual resident memory for your thread and you needn't care about the notional "real" size of the thread since we're 64-bit and address space is basically free.
- foderking 3y ago
  > (but of course named co_async and co_await, because C++ isn't the sort of language where you're allowed nice things, the beatings will continue until morale improves) LMAO
  - gpderetta 3y ago
    co_<whatever> has become a meme in the C++ community.
pkolaczk 3y ago
Stack memory is allocated lazily on Linux. I thought this benchmarks shows it pretty clearly. Otherwise 10k threads would be already in GBs territory, when in fact it is below 100 MB.

kaba0 3y ago

Since when is C the most performant? One can’t even write a performant generic vector with it. C++ is the de facto language used for hight performance.

remram 3y ago
"One can't even write a performant garbage-collected function in C++" You're describing features not performance.
- kaba0 3y ago
  So how do you write programs without data structures? You copy paste some implementation and modify it slightly at every use site? Possible optimizations depend on the semantics of the language — C++ can do just orders of magnitude more with for example a string data structure, than what is possible with a C “string”. So no, C is not the most performant language in general — de facto C++ is used for everything performance critical.
  - remram 3y ago
    > You copy paste some implementation and modify it slightly at every use site? Yes, or use macros. Not saying it's the most convenient, but that's an orthogonal concern. You wouldn't say hand-written assembly isn't performant, so your objection strikes me as wrong.