Isidor Zeuner wrote:
After coming along this ([1]) interesting link, I wonder what the suggested way of building a fully multicore-capable boost::asio tcp server on the linux platform would be. An obvious solution would be to listen on one interface/port per core and solve the problem at the application layer, but I'm looking for a fully transparent solution. I haven't looked more deeply into the boost::asio inner workings. Is it feasible to have modify epoll_reactor to distribute the file descriptors among multiple epoll threads?
Best regards,
Isidor Zeuner
[1] [http://stackoverflow.com/questions/1234750/c-socket-server-unable-to- saturate-cpu
Are you aware of the current performance initiative by Chris? http://article.gmane.org/gmane.comp.lib.boost.asio.user/3532 Cheers, Rutger