httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeff Trawick <trawi...@bellsouth.net>
Subject Re: Nagle
Date Thu, 14 Sep 2000 00:48:54 GMT
mja@trudge.engr.sgi.com (Mike Abbott) writes:

> No 2.0a6 listen socket has the Nagle algorithm turned off (TCP_NODELAY
> set), but every 1.3 listen socket does.  (Compare make_sock() in the
> two.)  Why the difference?
> -- 
> Michael J. Abbott        mja@sgi.com        http://reality.sgi.com/mja/

O.k., I guess I'm finally catching on to your point...

1.3: we disable nagle on the listening sockets *and* on the connected
     sockets...

2.0: we disable nagle only on the connected sockets

A possible optimization is to disable nagle on the listening sockets
(like 1.3) and (unlike 1.3) for the various platforms in which
connected sockets will inherit the nagle setting from the listening
socket we should *not* disable nagle again on the connected sockets.
(But for other platforms we'd still disable nagle on the connected
sockets.) 
And since your MPM doesn't have the call to disable nagle on connected
sockets, you're happy once we do it for listening sockets again.

So that is a bug? (snort)

But of course it is a cool thing to do, and it doesn't hurt Apache if
make_sock disables nagle but (at least at the present time) the
Apache-provided MPMs don't know when they don't need to disable
nagle...  

Do you have the trivial (I guess) change to 2.0's make_sock() already
tested? 

Thanks,

Jeff
-- 
Jeff Trawick | trawick@ibm.net | PGP public key at web site:
     http://www.geocities.com/SiliconValley/Park/9289/
          Born in Roswell... married an alien...

Mime
View raw message