On Mon, 23 Mar 2009 12:06:39 -0400, Michael DeHaan mdehaan@redhat.com wrote:
This is a development branch thing but it's related to Cobbler's plan moving forward, so it seems best to discuss here.
So ... there's been some good work going so far towards supporting some variants of Windows, though I see it may perhaps be somewhat of an evolutionary dead-end.
The web page for RIS-Linux in particular does not mention Vista or 2008. Further, I am much more interested in supporting Windows
virtually
than physically -- this should be natural with the increasing interest in virtualization and recent interoperability agreements with Microsoft.
For the devel branch (1.7), I'd rather we refocus our efforts into making sure the experience for Windows installs, virtually, is as good as possible.
The upside of this is that most of the infrastructure is already in place -- we can already do ISO based fullvirt installs ("cobbler image add" with the ISO file residing on NFS) and a next step is to teach it about a virtual floppy drive with the SIF answer file on the drive, so it can be fully scripted.
We already also have the "virt-clone" image type, for being able to take an existing disk image and repeatedly clone that image with koan, keeping the same source image on NFS. (The syntax here is "koan --image=foo --virt", just like with the ISO based installs for virt).
The goal here is to not invest too much effort in supporting dead-end deployment areas and writing code to cater to say, XP vs Vista vs 2003 vs 2008, but handle things generically, with answer files, and images, things we already do. On the plus side, there's also much less work in doing this and no additional dependencies or things to configure and set up.
Rather than physical deployments this encourages deploying Windows on Linux hosts, which makes the Windows machines easier to manage since you can run tools like libvirt and Func on the hosts. See http://www.redhat.com/promo/svvp/
Windows belongs running on virt. In general, for non-Linux OS's, we should also concentrate on virt.
--Michael
cobbler mailing list cobbler@lists.fedorahosted.org https://fedorahosted.org/mailman/listinfo/cobbler
Here are my thoughts on why this may be the correct long term goal, but bad for the short term (and by short term I mean the next 5 years). Small-Mid size corporations are quite often heterogenous, with many different OS'es running in the server room/data center. This is also true with Large companies, though you run into issues of momentum and territoriality there, so getting new build environments in is not an easy thing to do.
When I was working at a pretty good sized hosting company/Tier-1 ISP, one of the complaints I fielded against cobbler was that it was Red Hat specific, which at the time it very much was. In a company that deployed RH/Solaris/Windows with fairly equal regularity on real hardware, there would be no chance of replacing the in house build systems with cobbler.
Virtualization certainly has its place, however currently there are many pitfalls preventing its complete adoption (real-time systems in particular). As I stated above, I expect this to change within 5 years or so, but for now this decision would take away the ability of admins to use cobbler as the Swiss-army knife of build servers when deploying to bare metal systems. I also believe that if you can establish a beach head in the data center for cobbler, you will open avenues for adoption by demonstrating how easy it is to deploy virtually (no pun intended) as many OSes as are commonly used on many different platforms.
I'm obviously a bit biased, being the developer interested in supporting as many non-linux os'es as possible, although I certainly understand the RH viewpoint and the stategery involved in the decision. I won't be heart broken if we don't decide to support bare metal for most (if not all) non-linux os'es, but I do think it's a missed opportunity.