[Haskell-cafe] System calls and Haskell threads
I just read Kazu Yamamoto's article on a high performance web server in the latest Monad.Reader, and I came across a statement that doesn't sound correct to me. He says: When a user thread issues a system call, a context switch occurs. This means that all Haskell user threads stop, and instead the kernel is given the CPU time. Is this right? I thought that when a system call is made by a Haskell thread being run by a particular worker thread on a CPU, other runnable Haskell threads in the run queues of the HECs for other CPUs can continue running concurrently (provided we've run our Haskell program with multiple CPUs using the -Nx RTS argument). That's what I understood from the discussion of foreign calls in Runtime Support for Multicore Haskell. Andreas ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] System calls and Haskell threads
On Thu, Nov 3, 2011 at 8:35 AM, Andreas Voellmy andreas.voel...@gmail.comwrote: I just read Kazu Yamamoto's article on a high performance web server in the latest Monad.Reader, and I came across a statement that doesn't sound correct to me. He says: When a user thread issues a system call, a context switch occurs. This means that all Haskell user threads stop, and instead the kernel is given the CPU time. Is this right? I thought that when a system call is made by a Haskell thread being run by a particular worker thread on a CPU, other runnable Haskell threads in the run queues of the HECs for other CPUs can continue running concurrently (provided we've run our Haskell program with multiple CPUs using the -Nx RTS argument). That's what I understood from the discussion of foreign calls in Runtime Support for Multicore Haskell. That's correct. Blocking syscalls will not prevent other Haskell threads from running. IIRC it will block the OS thread used to run the Haskell thread making the blocking syscall, but the RTS always has one free OS thread (i.e. it will allocated more if needed) that it can use to continue running other Haskell threads with. Your best reference is probably Extending the Haskell Foreign Function Interface with Concurrency. -- Johan ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] System calls and Haskell threads
On Thu, Nov 3, 2011 at 8:35 AM, Andreas Voellmy andreas.voel...@gmail.comwrote: I just read Kazu Yamamoto's article on a high performance web server in the latest Monad.Reader, and I came across a statement that doesn't sound correct to me. He says: When a user thread issues a system call, a context switch occurs. This means that all Haskell user threads stop, and instead the kernel is given the CPU time. Is this right? It is correct in context. Mighttpd does not use the -Nx argument to create multiple OS threads, instead uses a `prefork` model that creates separate processes to balance user invocations. Using multiple processes instead of multiple Haskell threads avoids issues with garbage collection. Regards, Dave ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] System calls and Haskell threads
On Thu, Nov 3, 2011 at 10:22 AM, David Barbour dmbarb...@gmail.com wrote: It is correct in context. Mighttpd does not use the -Nx argument to create multiple OS threads, instead uses a `prefork` model that creates separate processes to balance user invocations. Using multiple processes instead of multiple Haskell threads avoids issues with garbage collection. I should rephrase that: using multiple processes in place of -Nx threads avoids issues with GC. Mighttpd does use multiple Haskell threads per Haskell process. ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
