So, I'm sure this is an age old dilemna but I'm going planning on doing something about it.

Image X is a gold image
Person A, B, C all want to use Image X for whatever purpose.
Proceed with making 3 copies of X.
Person A, B, C all finish customizing Image X.
Image X is saved back 3x as Image Xa, Xb, Xc.
This is how it's normally done.
You can go 1 step further and try using De-dupe/thin provisioning on the filer to try and provision these disks as just the differences from gold Image X, but that's a post process spike in I/O.

I'm looking for a program/driver that will basically fork the writes. So the only thing that separates Image Xa, Xb, & Xc from X are the discrete write operations performed during the customization process. I would like to capture either all the files that were written to or some how determine which have been written to (checksum the FAT?) and then batch process an export copy to an external disk.

I'm now left with Xa, Xb, Xc taking up ~2%*3 of Image X instead of 100%*3.

So before I try to write something that will inevidibly be a pain in the ass to maintain, is there something out there that would do this? Perhaps if I was booting over a network protocol and then modified the network driver to broadcast to multiple locations and trap the write operations? This could work since network I/O analysis speed >>> disk I/O

4/6/2009 3:45:18 PM

LVM snapshots

4/6/2009 4:58:17 PM

^^Hyper-V does exactly what you are asking for. What platform is this on?

4/6/2009 10:37:01 PM

currently ESX with all x86 OS's. I'd like it to be non-platform specific though. I think cache on write with etherboot might be the way to go

4/7/2009 8:18:48 AM

also, I'd like these deltas to be in a portable format, like raw files. this way I could pass around the "personal" copy of the image as a tar ball or something and do stuff with it out of band.

4/7/2009 8:40:54 AM

You can do all of this with Hyper-V.

Why does platform specificity matter at all with virtualization? What do you gain by being platform agnostic on the server end? Hyper-V supports a HELL of a lot more guest OS configuration than VMWare, and (apparently) has a lot more imaging management capabilities.

4/7/2009 6:34:49 PM

by platform agnostic, I mean outside of just x86, which hyper-V does NOT support. this feature set should be independent of the hypervisor as I'd like to be able to implement on any machine, physical or virtual, as a means of thinly provisioning disk. currently I'm implementing this at the filer level (netapp) but I hate being tied to a storage vendor and netapp isn't a very good one.

I've basically backed myself into the corner of doing this at the driver level and since the disk connection is likely coming over ethernet (trying to get away from fiber channel). I'll likely boot from iscsi or AoE and implement a copy on write algorithm that writes to another network attached disk.

What I want is something open and a standard. I don't want some bullshit bin that can only be opened by some emc tool, netapp tool, ms tool, etc. With this implementation, I'd have plain old files in a directory that could easily be tar'd, transported, examined, etc.

[Edited on April 8, 2009 at 11:33 AM. Reason : man, this really makes me think of that RFC article]

4/8/2009 11:26:38 AM

Quote :

"Why does platform specificity matter at all with virtualization? What do you gain by being platform agnostic on the server end? Hyper-V supports a HELL of a lot more guest OS configuration than VMWare, and (apparently) has a lot more imaging management capabilities."

yes, but if you've already made a significant investment in VMware, it makes no sense to migrate to hyper-v.

4/8/2009 12:35:11 PM