On Sat, 04 Oct 2008 17:16:04 +0200, Boris wrote:

[...]

...

I have a question regarding the system limit on a *NIX system for the maximum open file descriptors that a process may have open.

If you were using the boost asio library to design a server application how would you go about and extend the server to handle more than the 1024 file descriptor limit.

I think the 1024 file descriptor limit is only a problem if you use select(). Depending on your operating system it should be possible to raise the per-process limit. If you use then an I/O service based on poll() or epoll() I would assume this limit goes away.

Etienne, this link might also help you: http://www.kegel.com/c10k.html#limits.filehandles Boris

On Sat, 04 Oct 2008 18:04:10 +0200, Etienne Philip Pretorius wrote:

[...]So I guess, my question is how do I span multiple sockets/file descriptors across multiple processes by using boost libraries.

Boost.Process might help you: http://www.highscore.de/boost/process/ Boris

Hello Boris, On Sat, Oct 4, 2008 at 12:23 PM, Boris wrote:

Boost.Process might help you: http://www.highscore.de/boost/process/

boost.process looks very promising, just wondering are there plans to include it with boost 1.37. Regards, Sebastian

Thank you Boris, I have been using strace to track what is being called by the boost asio libraries, and in my 2.6 kernel it is epoll. getrlimit(RLIMIT_STACK, {rlim_cur=8192*1024, rlim_max=RLIM_INFINITY}) = 0 uname({sys="Linux", node="etienne-laptop", ...}) = 0 futex(0xb7f46674, 0x81 /* FUTEX_??? */, 2147483647) = 0 futex(0xb7f4b700, 0x81 /* FUTEX_??? */, 2147483647) = 0 brk(0) = 0x805c000 brk(0x807d000) = 0x807d000 epoll_create(20000) = 3 pipe([4, 5]) = 0 fcntl64(4, F_SETFL, O_RDONLY|O_NONBLOCK) = 0 fcntl64(5, F_SETFL, O_RDONLY|O_NONBLOCK) = 0 epoll_ctl(3, EPOLL_CTL_ADD, 4, {EPOLLIN|EPOLLERR, {u32=4, u64=4}}) = 0 socket(PF_INET6, SOCK_STREAM, IPPROTO_TCP) = 6 epoll_ctl(3, EPOLL_CTL_ADD, 6, {0, {u32=6, u64=6}}) = 0 setsockopt(6, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0 bind(6, {sa_family=AF_INET6, sin6_port=htons(1234), inet_pton(AF_INET6, "::", &sin6_addr), sin6_flowinfo=0, sin6_scope_id=0}, 28) = 0 listen(6, 128) = 0 ioctl(6, FIONBIO, [1]) = 0 accept(6, 0, NULL) = -1 EAGAIN (Resource temporarily unavailable) epoll_ctl(3, EPOLL_CTL_MOD, 6, {EPOLLIN|EPOLLERR|EPOLLHUP, {u32=6, u64=6}}) = 0 mmap2(NULL, 8392704, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb74ce000 mprotect(0xb74ce000, 4096, PROT_NONE) = 0 clone(child_stack=0xb7cce4b4, flags=CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|CLONE_SYSVSEM|CLONE_SETTLS|CLONE_PARENT_SETTID|CLONE_CHILD_CLEARTID, parent_tidptr=0xb7ccebd8, {entry_number:6, base_addr:0xb7cceb90, limit:1048575, seg_32bit:1, contents:0, read_exec_only:0, limit_in_pages:1, seg_not_present:0, useable:1}, child_tidptr=0xb7ccebd8) = 15388 rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0 rt_sigaction(SIGCHLD, NULL, {SIG_DFL}, 8) = 0 rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0 The limit unfortunately still applies. I am using a shell script to test this on the Ubuntu distro: for ((i=0;i<1024;i++)) do echo $i; `nc -n -w 1 127.0.0.1 1234 & `; done In the code I output an error when the accept failes: void server::accept( client* connection, const asio::error_code& ec ) { if(!ec) { client_.push_back(connection); }else{ //TODO: say why the connection failed delete connection; std::cout << ec.message() << std::endl; ::exit(0); } accept(); };//accept The returned error message is: Too many open files