check out template tree II ... Re: [Infrastructures] fai vs. ZENworks vs. m23

[Tobias Oetiker]

Hi,

this makes me want to promote template tree again ...

> I'm using FAI and isconf 2.X to manage about 20 Debian Sarge machines
> for a manufacturing test environment.  It has worked well for us.  For
> the next infrastructure I will definitely be using isconf 4.X.  I just
> have too much knowledge stored in makefiles right now to switch.

Here at ETH Zurich (EE-Department) we run

  196 Debian 3.1 ia32
   18 Debian 3.1 amd64
  148 Solaris 8
   10 Solaris 9
   59 Mac OS X 10.3

amongst these are 343 diskless clients.

We run Desktop Workstations, Server, Compute Clusters and Student
Labs. All configured from one high-level configuration file of 2395
lines ... this file is well structured and pulls together 359
configurable feature modules that make up the actual configuration
code. Each feature can exhibit an API that makes it possible
to use the same module in different configurations.

The configuration info gets converted to a cfengine.conf file for
application to the machines ... if we have the system create a
single cfengine.conf file describing all the config we apply then
the file has 1.2 million lines of cfengine code (this could be
optimized). A config for one server still comes in at 4979 lines of
cfengine code. A client is about 2864 lines.

The system we developed and use is called template tree II. You can
read all about it on

http://isg.ee.ethz.ch/tools/isgtc/index.cgi?page=module_pod;module=tetre2;pod=docs/overview.pod

currently you can get all the code but we have not yet published a
set of sample configuration files since it all tends to be rather
site specific (we wouldn't need configuration management
otherwise).

I mentioned diskless clients above. The creation of diskless
clients is fully automated too. It takes about 45 Seconds to build
a diskless client image. Images are built offline under a separate
mount point. Once an image is built the DHCP config gets adjusted
such that when the client boots the next time, it will
automatically pick up the new image. This makes down-times for
'upgrades' a matter of seconds ... We don't actually ever patch
diskless clients, since we just rebuild them from scratch whenever
a new kernel is deployed or we add in some more Solaris patches ...
Currently we do diskless Solaris and Debian, but OSX 10.4 looks very
promising.

http://isg.ee.ethz.ch/tools/isgtc/index.cgi?page=module_pod;module=disklessmgr2;pod=disklessmgr2

cheers
tobi

-- 
 ______    __   _
/_  __/_  / /  (_) Oetiker @ ISG.EE, ETL F24.2, ETH, CH-8092 Zurich
 / // _ \/ _ \/ /  System Manager, Time Lord, Coder, Designer, Coach
/_/ \.__/_.__/_/   http://people.ee.ethz.ch/oetiker +41(0)44-632-5286