[Savannah-hackers] Re: cpuload on savannah

[Mathieu Roy]

"Jaime E. Villate" <address@hidden> said:

> On Thu, Mar 13, 2003 at 05:25:32PM +0100, address@hidden wrote:
> > 
> > after some investigations, i didn't find anything wrong that some zombies
> > of cvs (about 30). after the kill of all of them, the cpuload is about 80
> > (lot of cvs, viewcvs and apache-ssl process).
> > 
> > savannah remains very very slow (working on subversions if very
> > fastidious). 
> > 
> > I don't know if there is a plan  of upgrading savannah, but i can't do
> > anything more now ..
> 
> Hi,
> Thanks for your efforts. It seems to me that we have a lot of anoncvs and
> viewcvs processes because as the server has had some problems lately, users
> try desperately to access the cvs repositories. A lot of those processes might
> have been started by scripts automated by the users; that is very hard to
> control. The only solution I see is to make an upgrade to a more powerful
> architecture; a faster CPU with more RAM. And if the number of hosted projects
> continues to increase at the current rate, we will soon need a cluster to keep
> Savannah running.

I do not think we should make any decision until we get less than
25000 https request for mldonkey files - which is absolutely not
usual. 

We cannot make a decent evaluation of what is really needed which 70
load average, while usual cronjob takes hours to be done - and so
dramatically contribute the load rising.

After checking and rechecking, the main problem is obviously the
number of httpd process and that the big difference.

Aside from that, I think that hosting CVS on an another box would be a
better solution than buying a new computer.

And about buying material, only RAM seemed really needed before the
mldonkey issue.



-- 
Mathieu Roy
 
 << Profile  << http://savannah.gnu.org/users/yeupou <<
 >> Homepage >> http://yeupou.coleumes.org           >>
 << GPG Key  << http://stock.coleumes.org/gpg        <<