httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From John Heidemann <jo...@ISI.EDU>
Subject Re: performance ``bugs'' in apache-1.1b4's P-HTTP
Date Thu, 11 Jul 1996 17:05:19 GMT
On Wed, 10 Jul 1996 19:44:47 PDT, Cliff Skolnick wrote: 
>Using writev() instead of a whole series of write() calls is a big
>perfomance win.  Sun's internet engineering guys have been trying to
>convince me this is a really good thing.  They have the data to back it
>up, but this is not a small effort.
>Turning off the nagle algorithm is nto a good idea.  We ran with this off
>at Organic for a while and we ended up increasing our packet count
>measurably and burning lots more of our lines because of this.  From my
>experience this is not a good thing, and the TCP/IP gurus who say use
>writev() very strongly also say very strongly not to do this.

These experiences are true for interactive traffic with small writes.
(For example, Nagle is very important for telnet traffic.)

I don't believe they apply to Apache.  Apache nearly always writes
data in 4KB units.  (Nearly all Apache data goes through the buffered
IO routines.)  Since the only data to send is in a single contiguous
buffer, writev has no advantage over write.  Since the data is written
in relatively large chunks, Apache nearly always writes full-size
segments and the Nagle can't reduce packet count.

   -John Heidemann

View raw message