There have been a lot of suggestions (on here and other sites) to run puppet with locally-rsync-ed (rsunk?) copies of manifests, but there are a few things which won't work if you do this, unfortunately. Most importantly is the `storedconfigs' which (afaict) require the puppet server to work.
This means that you lose a large amount of the power of using puppet, by which you can use configurations across machines to do things like collect up all the services you run on a set of machines and generate a nagios config, or firewall config, or whatever. Without using stored configs for this I assume it's possible but will require more explicit configuration rather than the rather more elegant solution provided when using a puppet server.
Side note : I've used puppet on a fairly small scale of up to ~50 machines, and just started using it for VMs, and it's pretty straightforward to integrate into a bootstrap install to get ruby and puppet installed so that you can use it to install all the rest of the dependencies. But of course, the most use is for changes later on rather than at install-time when there are already a huge number of tools to set up or image machines or whatever.
Side side note : I've not used Chef to compare this with.
This means that you lose a large amount of the power of using puppet, by which you can use configurations across machines to do things like collect up all the services you run on a set of machines and generate a nagios config, or firewall config, or whatever. Without using stored configs for this I assume it's possible but will require more explicit configuration rather than the rather more elegant solution provided when using a puppet server.
Side note : I've used puppet on a fairly small scale of up to ~50 machines, and just started using it for VMs, and it's pretty straightforward to integrate into a bootstrap install to get ruby and puppet installed so that you can use it to install all the rest of the dependencies. But of course, the most use is for changes later on rather than at install-time when there are already a huge number of tools to set up or image machines or whatever.
Side side note : I've not used Chef to compare this with.