The articles appear to be referring to the footprint of the tool rather than the scalability.  The aim for us is to support O(100K) clients and we can accept some hardware investment in a scale-out architecture to achieve this. 

The approach is not a 'developer' one but an infrastructure manager looking at total cost of ownership which includes more than hardware costs. We've certainly found major productivity gains from tools such as PuppetDB and Heira along with manifests from the puppetforge such as the OpenStack one.

As you say, the most important thing is to do configuration management, the tool choice is secondary and different sites will have different needs.

Tim

> -----Original Message-----
> From: Natxo Asenjo [mailto:[log in to unmask]]
> Sent: 22 February 2013 23:18
> To: Tim Bell
> Cc: Yasha Karant; [log in to unmask]
> Subject: Re: puppet
> 
> On Fri, Feb 22, 2013 at 6:54 PM, Tim Bell <[log in to unmask]> wrote:
> >
> > I'm not sure where the statement that Puppet does not scale well comes
> > from...
> 
> it is pretty well documented by Jarle Bjørgeengen for ;login (usenix):
> 
> https://www.usenix.org/publications/login/february-2010-volume-35-number-
> 1/puppet-and-cfengine-compared-time-and-resource
> 
> This one is more recent:
> 
> http://www.blogcompiler.com/2012/09/30/scalability-of-cfengine-and-puppet-
> 2/
> 
> And another happy puppet user:
> 
> http://my.opera.com/marcomarongiu/blog/2012/06/17/why-i-gave-up-puppet-
> and-chose-cfengine-3
> 
> > You do need to have enough puppet masters and architect it according
> > to the best practises but the books tell you how to do this.
> 
> exactly, a developer solution to every problem: throw more hardware at it,
> hardware is cheap ;-)
> 
> Specially in virtual environments puppet is a resource hog if you compare it to
> cfengine. I have exactly one cfengine policy host controlling nearly 400 hosts.
> Its load is nearly always 0. This is quite normal. The policy host has 3 different
> sets of policies: test, staging and production. This is a vm with a grand 256Mb
> ram and 1 vcpu. This host is our kickstart/fai installation server as well.
> 
> SPOF you may think. Not really, hosts cache their policies, so when the policy
> master needs patching, whatever, we just do it. The hosts apply their cached
> policies.
> 
> I am pretty much sure I could not run the same load with the same hardware
> with puppet (actually, I know, I tested it).
> 
> O well, in the end as long as people use something sane I am happy.
> --
> natxo