From: spam@reppisch.de   
      
   Hi Ng,   
      
      
   i've further investigated on this latency problem.   
   The test-client has 10 threads each of which opening a connection to the   
   server and doing blocking io (send/recv pingpong).   
      
   It seems that 2 out of 100k messages over all threads were about 6   
   seconds "in transit".   
   Time was measured between message assembly & send() to dequeueing from   
   the recveive buffer.   
      
   Meanwhile other sockets did perform properly. (typical 0,5 ms in transit).   
   Any ideas on that? Packet loss? Socket observing bug?   
      
   Should i take control of the stack's behaviour about sending data in   
   more "real-Time"?   
   I've set the SOL_SOCKET, SO_SNDBUF/SO_RCVBUF Buffers to 65k for   
   negotiating the "bigger window extension".   
   Maybe i could try to disable Nagle, but in my opinion the stack should   
   not have to be tweaked.   
      
   Regards,   
      
     Michael   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)