|
Home > Archive > Unix True 64 > March 2004 > Impact of the number of TCP/IP connections on the same port on Apache web server perfo
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
Impact of the number of TCP/IP connections on the same port on Apache web server perfo
|
|
| Geert Van Pamel 2004-02-28, 12:34 pm |
| I have some practical questions about Apache web server network performance
on Tru64 UNIX 4.0F
- I am running an intranet Apache web server with php scripts and Oracle
DBI.
- I get typically 20 hits /s (mix of pure html, JPG/GIF, others are PHP
scripts)
- I have e.g. 100 open connections on one and the same TCP/IP port.
Question [1]
Could I have a better performance if I used multiple ports with multiple
Listen instructions in httpd.conf instead of listening to only 1 port 80
vi /usr/local/apache/conf/httpd.conf
....
Listen 2001
Listen 2002
Listen 2003
Listen 2004
Listen 2005
....
Could the performance be better when having 5 x 20 active connections
instead of 1 x 100 ?
Question [2]
Does there exist (kernel or other) parameters that I can tune for better
performance.
Question [3]
Does anybody have experience with persistent database connections? The
problem is that creating an Oracle process every (?) second generates up to
10 % of (unproductive) system CPU... and extra delays in processing the
queries of course.
Thanks,
| |
| Rick Jones 2004-03-02, 8:35 pm |
| There are some old SPECweb99 and SPECweb96 disclosures for Tru64 at
http://www.spec.org/ . They are likely all for Zeus, rather than
Apache, but perhaps some of the settings might be germane.
I cannot speak directly to Tru64, but on HP-UX, if there were only 20
hits per second, I don't think that splitting things across multiple
listen endpoints would make much of a difference. At least not so far
as networking is concerned.
BTW, as far as having 100 connections on one and the same port, that
is only the local port. The TCP connections themselves are "named"
with the four-tuple of local/remote port, local/remote IP, and about
the only place I can think of in a TCP/IP stack where the one
well-known port could be a bottleneck is during connection
establishment as that is the only time things feed just through the
listen socket.
rick jones
--
Process shall set you free from the need for rational thought.
these opinions are mine, all mine; HP might not want them anyway... 
feel free to post, OR email to raj in cup.hp.com but NOT BOTH...
|
|
|
|
|