On Wed, May 04, 2005 at 11:02:31AM -0500, Sudong Lee wrote:
> 
> Hi Juergen

Hello,

> I been reading most of your writings and they were very helpful.
> 
> I remember you had some download links for mathopd that was compiled with 
> dietlibc. You had 2 donwnload links and now I only see one.
> 
> What happen to the php part??

I guess you mean the cgi-fcgi/spawn-cgi utilities. I removed them,
because I thought that they are a little bit too specific and not
really mathopd related. They are still available at [1], maybe I
should add them again.

> Can you also write a how to on how you compiled mathopd with dietlibc.

ok, I can do that, though it's not a big issue to compile mathopd
against dietlibc, it works without patching.

> Do you think compiling mathopd with dietlibc gives significant improvement?

hmm, that's a good question. The main advantage is surely that the
binary is statically linked and thus depends on nothing. It should
run on every Linux distribution if you have an i586 or better cpu.
And that's the main reason, why I put the link to the binary into
the wiki.

The other advantage is the very small memory footprint of dietlibc
binaries.  With NumConnections = 1 and default values for the 
buffers I got the following:
         
              VSZ    RSZ
  ----------------------- 
  glibc     1732K   648K
  dietlibc   256K   140K

But that's not important with mathopd, because most of the memory
goes into the buffers if you have a usable setup.

Another point is overall performance, here is dietlibc ahead too.
A short test with apache's ab and a static web page of around 2k
with 10 concurrency connections gave me 2000 to 2600 requests per
second, which is a nice improvement. But ab is only a test ....

Greetings
Juergen


[1] http://jue.li/crux/dietlibc/fcgi-utils-1.0-1d.pkg.tar.gz


-- 
Juergen Daubert  |  mailto:[EMAIL PROTECTED]  
Korb, Germany    |  http://jue.li/crux

Reply via email to