On Mon, Sep 5, 2016 at 4:36 AM, Dan <dane...@hotmail.com> wrote:
> Hi,
>
>
> I am trying to create a Julia server instance by using ZMQ with the
> Req/Rep pattern. My Julia code for the Rep-server looks like this:
>
> module JServer
>
> # import zmq
> using ZMQ
> import Base.gc
>
> # create context
> ctx=Context();
> # create socket
> socket=Socket(ctx, REP);
> # msg
> msg = 0;
> # repeatedly collect garbage every second
> timer = Timer(gc, 1, 1);
>
> function connect(host, port)
>
> # connect
> ZMQ.connect(socket, string("tcp://", host, ":", port));
> println(string("Julia server connected to ", host, " on port ", port));
> end
>
>
>
> function startServer()
> # notify
> println("Waiting for request...")
> i = 1
> while i <= 2
> # invoke garbage collection
> # gc();
> # Poll For message
> msg = receive();
> println("Message received");
> # Reply
> send();
> println("Message acknowledged");
> end
> end
>
>
>
> function receive()
> msg = ZMQ.recv(socket);
> end
>
>
>
> function send()
> ZMQ.send(socket, Message("test request"));
> end
>
>
>
> function gc(event)
> gc();
> end
>
> end
>
> I can successfully start the server by runnung this script (indicating
> host and port via ARGS):
>
> # include server module
> include(string(dirname(Base.source_path()), "\JServer.jl"));
> # connect server
> JServer.connect(ARGS[1], ARGS[2]);
> # start server
> JServer.startServer();
>
> My tests seem to be odd, however: Whenever I send large messages from
> client, i.e. ~1MB per message, Julia seems to properly keep tracking about
> memory usage: The overall process memory remains constant. But when I am
> sending small messaged (i.e. 10k per message) Julia memory usage is growing
> permanenlty. In addition response time slows down over my requests. This
> does not happen if I explicitly invoke the garbage collector inside the
> while loop, but this would slow down the replies significantly.
>
>
> I am using Julia 0.4.6.
>
>
>
Try 0.5.
> Is my code ok? Is there a callback solution in order to be able to avoid
> the while loop?
>
