On 01/28/2013 04:04 PM, Rai, Neeraj wrote:
I had tried interprocess communication before and found it to be slow (12 sec vs 14ms). Somewhere in the docs, there was a mention that it has latency of 2ms. However, if you can point me to samples or provide other advise that make it as fast as osl::socket, I'd be happy to switch back to it.
"Generic" UNO IPC with all its bells and whistles easily introduces more overhead than any solution tailored to one specific use-case, but "12 sec vs 14ms" surely sounds broken. Hard to tell what's gone wrong there from a high-level perspective.
Q2. Instead of using osl::AcceptorSocket and osl::ConnectorSocket, I tried to use #include <sys/socket.h> int socket = (PF_INET, SOCK_STREAM, 0); bind(socket, addr, addrlen); listen (socket, 5); This would allow me use scalable epoll but I had trouble compiling it.
There should be no general reason this should not work. (It would mean the code is platform-specific, of course.)
Q3. I would like to understand more about how these threads interact with connection::read/write. Is there an extra thread created for me per connection ? write would be called in my user thread. Is the data queued for writer thread ? read is called in my user thread. Is it listening on some UNO queue to which the dedicated reader thread queues data ? Or did you mean that I should create a dedicated thread per connection because read/write are blocking calls ?
Reading/writing are blocking operations at the OSL socket level. Stephan _______________________________________________ LibreOffice mailing list LibreOffice@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice