On Friday 01 October 2004 15:57, Henrik Nordstrom wrote:
> On Fri, 1 Oct 2004, Jamie Lokier wrote:
> > Do you instead mean that the server doesn't reassemble _segments_?
> > Even when those segments are transmitted consecutively?
>
> Only speculating here, but more likely the application does not
> handle messages split in multiple segments and expects the whole
> message to be returned in a single read call..
I'm 100% convinced that the application server, on the other side, is
broken in exactly that way. It's with a bank. Going to their change
control system is horrendous and slow. If I can work around the
problem in less than six months, I'm happy :)
I didn't mean to imply Linux is broken. The other side is broken. I
wanted to know why Linux splits the packet into 1/2 the window size,
when sending, when the window size is less than the MSS, to try and
work around the problem.
There is more than one way around the problem:
1. Send raw IP packets. Lots of work, for very little payback.
2. Put a router inbetween that rewrites the advertised RWIN.
Not even going to go there.
3. Take 2 lines of source out of the kernel. Works. Might break
other comms.
If I know /why/ the upper bound is half the window size, maybe I can
patch it so nothing breaks.
> Have seen too many such applications, sometimes from very big
> vendors.
I've seen that before, as well, in various places.
> Regards
> Henrik
> -
> : send the line "unsubscribe linux-net"
> in the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Regards,
Berend
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
