Re: NAGLE commented

Andi Gutmans (andi@vipe.technion.ac.il)
Sat, 28 Jun 1997 01:42:52


At 23:29 27/06/97 +0100, Alan Cox wrote:
>> Is there a reason that NAGLE is commented in net/ipv4/Config.in? Is the
>> disabling of it broken?
>
>Disabling Nagle is not a valid option for a tcp stacks default behaviour,
>therefore its not allowed.

Well the reason I wanted to disable it, is that I'm suffering a
performance problem running ppp via a binary telnet. The performance
problem isn't really in throughput as it is in response time. Instead of
having a ~200ms ping I get ~350ms where the terminal servers are 1.8ms
away from the Linux box actually running pppd. I'm not escaping any
characters as it's a binary telnet connection. Someone told me there could
be two reasons for this:
a) The overhead of the telnet is causing this and there is no way around it.
b) The nagle causes this problem. So this is why I wanted to disable nagle.
What I did now is change telnetd to call (void) setsockopt(0, SOL_TCP,
TCP_NODELAY, NULL, (int)NULL);
but my knowledge of sockets is very small so I'm not sure this is the right
way to do it.

do you or anyone else have any idea if I can improve this "response time"
problem? (It has nothing to do with the Linux box's load as it really isn't
loaded).

Thanks, I'd appreciate any further insight.

Andi

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Andi Gutmans - Computer Science, Technion, Israel
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Web: http://andi.il.eu.org/ ICQ: 1568297 PGP: finger andi@vipe.technion.ac.il
KeyID 62F5661D fingerprint = 1A 87 A2 10 2F EF EF AB 47 E0 4D 42 F5 B5 49 AD