From: kuku@physik.rwth-aachen.de   
      
   Rick Jones  wrote:   
   > Christoph Kukulies  wrote:   
   > > A question about packet processing speed:   
      
   > > I have a server and a client which communicate via a socket (ipv4,   
   > > AF_INET).  The client sends a connect. The server does an accept and   
   > > waits for further messages from the client on that socket (using   
   > > select()).   
      
   > > The design of the higher level protocol is the following way:   
      
      
   > > client                                        server   
      
      
   > > send(REQUEST) --------------------------->    recv(REQUEST)   
   > > recv(DATA)    <--------------------------     send(DATA or ERROR CODE)   
   > > recv(END)     <--------------------------     send(END)   
      
   > > Now it turns out that the doubly sending of packets makes the whole   
   > > communication very slow and I have no idea why at the moment.   
   > > Each transaction in the above way takes 0.2 seconds and calculatiing   
   > > alone from the wire speed (100Mbit/s) for 256 bytes I would expect   
   > > smaller transmission times. Lowering the packet size down to 70   
   > > doesn't have any impact on the time I'm achieving.   
   > > But when I remove the sending of the END packet, I'm getting a   
   > > drastic performance increase, that is the transaction goes down to   
   > > 600 in 1 second.  A 600fold performance increase.   
      
   > Some boilerplate likely related to what you are experiencing:   
      
   A very valuable piece of information follows.   
   ....   
      
   Thanks, Rick. This was very helpful information and the solution   
   comes clear to me. Thanks for the other two posts.   
   --   
   Chris Christoph P. U. Kukulies kukulies (at) rwth-aachen.de   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)