Hello,

I am working on a project for a local library. They have about 30 systems that they will be running, and drives need to be cleaned every night (i.e. a reboot needs to happen and a fresh os needs to be installed). Does anyone have any suggestions for the best way to go about this?

I would prefer to use a centralized image for all of the client systems (rather than an image on each of the local discs). Is there a way to boot the system, fetch the image from a server, overwrite the / partition with it, and boot into that fresh os?

I've looked briefly at clonezilla as a potential solution. Any thoughts on that?

I've also looked at PXE, but the cards in these machines don't appear to support PXE boot? Is there a software-equivalent to PXE?

Thanks for any assistance.

You may also want to check g4u, http://www.feyrer.de/g4u/

why not a live image on a partition....


.

Unless the ethernet cards are weird, you can PXE boot. They may not have a PXE boot rom, but that still leaves the possibility of a PXE boot floppy. That's what I use on my older computers (Pentium, Pentium II, Pentium III).

What are the hardware specs on the computers in question?

Personally, I'd get rid of the hard drives altogether and PXE boot. One possibility is to use Knoppix or a Knoppix derivative so even the server doesn't need a hard drive. Knoppix has the option to be a diskless boot server; all of the PXE boot workstations boot up as if they were booting up Knoppix from a local CD.

I'd go for a more sophisticated setup, though. I don't know if there's a neat turn-key solution, but my solution would be a custom netboot initramfs image with scripts to unionfs an nfs root partition READONLY with a read/write local tmpfs ramdisk. This has a couple neat benefits:

1) Any number of client workstations can simultaneously use the same root partition (since it's mounted readonly).

and

2) Every time a client is rebooted, all changes to the OS are inherently wiped out. A tmpfs ramdisk is inherently non-persistent.

Setting up one workstation to boot up using a standard initramfs image (that mounts the nfs root partition read/write) lets that workstation be the "master". The "master" workstation can then be used to maintain software updates and customizing the user interface.

I don't see any reason for reinstalling the OS nightly. You can just mount the /home partition on a mem disk and it will disappear when there's a reboot.

If you really want a reinstall (it's going to kill your hard drive), you can use a floppy the LTSP project uses, etherboot.org. There are several cluster technologies which fit your bill, but I still recommend a mem disk.

Don't do it!

What most library systems do is not this, exactly. They give individual users a data area (which you can call home, but this is also done on windows systems) and that data area is wiped on either log on or log off. Now wiping that overnight wouldn't be a bad option either, but installing 'a fresh os' seems like overkill, and not only that creates the possibilty of extra problems.

I'm guessing that you probably don't want users to be able to store, e.g., UI config changes, which also argues for a complete wipe of home, but if you did want users to be able to store those, you'd only do a partial wipe. If going down this (IMHO more friendly) route, my feeling is that you need to give the library staff a 'wipe out user changes and restore defaults' button, or users will contuinually be finding interesting ways to make the system unworkable.

I'm surprised that PXE boot isn't supported at all. Sometimes the ethernet cards have a rom socket and you may need to populate that for PXE to work (given the price of ethernet cards it is not a surprise to find out that the rom isn't populated by default). There is also another system comparable to pxe boot, the name of which escapes me at the nmoment - have a look at the ltsp site.

One other thing to check - you don't want 'kiosk mode' do you?