PS. I should have mentioned that each worker maintains a socket *to* and *from* its partner. I have a situation where I have two (or more) nodes that need to periodically exchange data in a symmetric fashion - i.e. one needs to send the other a buffer, and vice versa. Currently I implement this as two async operations: for(neighboring nodes) { async_write(...) async_read(...) } m_ioService.run() There exists a tcp socket connection between each of the neighbors. The handlers associated with both read and write merely do error checking. Is this okay, or is there a issue with this approach? In general, it seems to be working fine, but someone brought a bug to my attention whereby they were getting "Transport endpoint not connected." errors happening (under a heavily loaded network, if that makes any difference, but I'm not using any timeouts). Thanks! Dan Eaton _______________________________________________ Boost-users mailing list Boost-us...http://groups.google.com/groups/unlock?_done=/group/boost-list/browse_thread/thread/b2f660881e76b428&msg=e93094ac386efd4d@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users