FreeBSD: stressing 4.2 and Apache with ab (fwd)

L. Cranswick L.M.D.Cranswick at dl.ac.uk
Mon Jan 1 05:14:37 GMT 2001


> > Do you have the parameters you have tweaked to make a
> > "robust" kernel?  Can you post the relevant parts of
> > the config files that could also be included on the web?
> 
> The only significant change from the GENERIC kernal configuration file
> is:
> 
> maxusers	512

Summary so far (Redhat 7.0 vs FreeBSD 4.2) (with a resulting query):

Hard to read Tables of results at :

 http://www.ccp14.ac.uk/solution/bsdunix/apache_comparison.html#testresults

------

Given anything can look good when compared to itself, I also
have installed Redhat 7.0 on the same PC and being going
through a near identical procedure:

1.  Using ab (apachebench) and apache 1.3.14 as a test:
  "Out of the box" Redhat 7.0 craps all over "out of the box" 
  FreeBSD 4.2 (due to default maxfile=16 setting).  Default FreeBSD
  looks totally woeful when compared in this way to
  Redhat Linux. 

2.  After recompiling the FreeBSD kernel with maxfiles=512 -
   on low to "high" (though not extremely high) loads, 
   default Redhat out performs FreeBSD
   in terms of speed (though the "loads" given out by Redhat are
   more variable - and at higher loads Redhat always seems to 
   registers a temporary zombie apache process which it then 
   gets rid of after it calms down from the test)

------------------------------------------------------

But just as things are getting interesting at very high loads (c => 1000 )
I get the message from AB when running on FreeBSD (maxfiles = 512) 
  
/usr/local/apache/bin/ab -n 1025 -c 1025 http://ccp14vig1.dl.ac.uk/small.html

  "Test aborted after 10 failures: Address family not supported 
   by protocol family:"

(no problems with -n 1024 -c 1024 )

Thus the Query:

Is there another tweak to get around this?  Or is this a hard limit?

--------

("default" Redhat barfs at ~ c=1010 due to "socket: Too many open files"
At higher loads for Redhat, "top" registers a temporary zombie apache
process which it then gets rid of after it calms down from the test(?).
(don't know why as yet?)
I think I have found multiple webpages to help me with the Redhat
limitations due to the Mindcraft - Linux-apache/Microsoft-IIS comparisons
and fall-out:
   How Mindcraft could have made a better Linux file server - Tuning Linux 
     http://www.linuxworld.com/linuxworld/lw-1999-04/lw-04-mindcraft.html
     http://www.kegel.com/mindcraft_redux.html#tcp3)

------

At these elevated levels c=1000 - FreeBSD starts to outperform
Linux on small HTML files - and I would still like to see where the 
the real limits are.


/usr/local/apache/bin/ab -n 1000 -c 1000 http://ccp14vig1.dl.ac.uk/small.html
                                                                (1.5 kB file)

Redhat 7.0:                - 4.3 seconds total,   228 requests per second
               Load ~ .7

FreeBSD (maxfiles = 512)   - 1.7 seconds total,   570 requests per second.
               Load ~ .3

======

/usr/local/apache/bin/ab -n 10000 -c 1000 http://ccp14vig1.dl.ac.uk/small.html
                                                                (1.5 kB file)

Redhat 7.0:                - 17 seconds total,    587 requests per second
               Load ~ .8
  
FreeBSD (maxfiles = 512)   - 15 seconds total,    667 requests per second.
               Load ~ .8

======

/usr/local/apache/bin/ab -n 1000 -c 1000 http://ccp14vig1.dl.ac.uk/large.html
                                                                (658 kB file)

Redhat 7.0:                - 35 seconds total,     28 requests per second
               Load varies - .6 to 5.6

FreeBSD (maxfiles = 512)   - 45 seconds total,     23 requests per second.
               Load ~ 1 (reproducable)

------------------------------------------------------

Cheers,

Lachlan.


-- 
Lachlan M. D. Cranswick

Collaborative Computational Project No 14 (CCP14)
    for Single Crystal and Powder Diffraction
Daresbury Laboratory, Warrington, WA4 4AD U.K
Tel: +44-1925-603703  Fax: +44-1925-603124
E-mail: l.cranswick at dl.ac.uk  Ext: 3703  Room C14
                           http://www.ccp14.ac.uk





More information about the Ukfreebsd mailing list