httpd-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stephane Bailliez <sbaill...@apache.org>
Subject [users@httpd] Apache 1.3.x performance config.
Date Tue, 15 Jun 2004 12:26:30 GMT
I have a client complaining about an endless CLOSE_WAIT (ie the page is 
apparently sent to the browser but not closed, so the browser display a 
blank page and wait endlessly.. doing another request seems to solve the 
problem). This is not fully evaluated and confirmed yet as they could 
not isolate nor replay the problem.

I had a quick look at the infrastructure (as I conveniently don't have 
access to it) and we noticed a couple of issues. Lots of sockets in 
FIN_WAIT2 and running out of file descriptors in mod_gzip)

I could not have information about the fd limit (rlim_fd_max and cur, 
but apparently 4096 ?) tcp_fin_wait2_timeout was set to get rid of the 
FIN_WAIT2 (and possibly helping fd issues, but did not have evaluation 
feedback). I could not run lsof to figure out what was using that much fds.

Environment:
Solaris 8 on Sun Fire 280, 8GB RAM
Max users: 25000
Concurrent users: 15
Requests: 4000 req/s
Used CPU: ~40%
Apache 1.3.27, mod_gzip, mod_expires (1 month for images, js, ...)

Note that without mod_gzip the bandwith is said not to be enough to 
serve all users (do not have figures)

Those are the numbers I have been given (when they have been given), 
consider them with a ton of salt. The max number of users is valid 
though, when I was there, they told me the current load at that time was 
about 500 req/s

The 2 apache boxes are front end to 2 Sun Fire 880, Solaris 8, 32GB RAM 
(8 Weblogic instances with 1GB JVM heap each)

Apache 1.3.27 relevant configuration snippet:

MaxClients		2000
MaxRequestPerChild	1000
MinSpareServers		2000
MaxSpareServers		2000
StartServers		2000

KeepAlive		On
KeepAliveTimeout	2
MaxKeepAliveRequests	0

What I'm concerned about is the consequences of setting such a high 
value on the number of Client and Servers (2000). From what they told 
me, they changed the HARD_SERVER_LIMIT to 10000 (I could not have access 
to the machine so could not check the result of httpd -V)

I find this quite an extremely high value, so I would appreciate any 
feedback about the coherence of such value considering the load and what 
is your experience with it, if you could give me figures about a given 
use cases that would be much appreciated. (I think however that IBM 
ships its Apache version with 4096)

Once again with such lack of information on the production machine, I 
can hardly do some effective work except guessing and throwing out 
ideas, so I know you should be in the same boat, but if you have hi-load 
configuration experience, you're welcome to share your known collatoral 
damages issues. :)

Cheers,

Stephane


---------------------------------------------------------------------
The official User-To-User support forum of the Apache HTTP Server Project.
See <URL:http://httpd.apache.org/userslist.html> for more info.
To unsubscribe, e-mail: users-unsubscribe@httpd.apache.org
   "   from the digest: users-digest-unsubscribe@httpd.apache.org
For additional commands, e-mail: users-help@httpd.apache.org


Mime
View raw message