Re: Sockets Pipe

[Scott Klement <klemscot@xxxxxxxxxxxx>]

I discussed this with Donnie off-list, and it turns out that his DR guy
had changed the default send/receive buffer sizes to 1024000 bytes.

Since his CPU is much faster than his network, his program was able to
quickly fill up that 1mb buffer.   Then, it took over 90 seconds for the
program on the other end to empty that buffer out, and that's why his
program was timing out.

His solution was to use setsockopt() to set the socket buffer sizes to
8k in his program, so that when other people make "optimizations" to the
network settings, it doesn't cause this problem.

On Mon, 10 Mar 2003 djobe@xxxxxxxxxxxxxxx wrote:

> How do you determine how large the sockets pipe is?
>
> I am using the select() with a 90 second time-out before a send() to
> determine if the socket is available to be written to.  Currently, the
> server program seems to be flooding the socket and timing out.  The
> application is a download of customer pricing and is sending a lot of data
> (23804 records * 60 bytes).  If I watch the batch program with wrkactjob it
> is in a run status, then selw, then it times out.  The client side does not
> have enough time to read all the data on the socket before the server side
> stops after the time-out.  The odd thing is the process used to work fine
> on this machine.  It crashed hard about a week ago and now my sockets
> process no longer works.
>
> If I point the client to another machine, it works just fine.  When I watch
> the batch program in wrkactjob, it varies between run and selw.  The client
> is able to read all the data from the socket and complete normally.  This
> leads me be believe there is a buffer (?) definition that controls the
> amount of data that can be placed on a socket.
>