6d66fa5e   
   From: rick.jones2@hp.com   
      
   > > Every TCP socket has a TX buffer. See the tcp(7) man page and   
   > > /proc/sys/net/ipv4/tcp_wmem. Seems to be 16kb by default on my system.   
   > > It grows when you send() and shrinks when the other side ACKs that it   
   > > has received part of the data.   
      
   I thought that in Linux SO_SNDBUF grew when the receiver's advertised   
   window grew and the sending cwnd grew as well.  I guess there is also   
   a "and is the application trying to use as much as it can" component   
   to the heuristic.   
      
   > > With netstat -t to see the size of the TX buffer, and with tcpdump to   
   > > see how well TCP is draining it.   
      
   Definitely take the tcpdump trace.  Perhaps in conjunction with a   
   timestamped system call trace of the application making the send()   
   calls.   
      
   Does smell like perhaps a packet loss issue.   
      
   rick jones   
   --   
   A: Because it fouls the order in which people normally read text.   
   Q: Why is top-posting such a bad thing?   
   A: Top-posting.   
   Q: What is the most annoying thing on usenet and in e-mail?   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)