Re: [SAGE] OS as an Appliance

["Cyrus Vesuna" <cyrus.sage@gmail.com>]

On 12/9/06, Rodrick Brown <rodrick.brown@gmail.com> wrote:
<snip>
> I basically want to build tiny, appliance like OS images where my
> servers will boot this image via PXE/BOOTP/DHCP/TFTP over a network
> then nfs mount our application binaries needed for that particular
> host. We sort of do this today with a few core tools and applications
> which are required on all hosts. Users home directories are also nfs
> mounted via automount.
>

Is this the only acceptable methodology? I see you are a Netapp user
and they (allegedly :-) ) have this
http://www.netapp.com/go/techontap/tot-march2006/0306tot_kilo.html and
http://www.netapp.com/go/techontap/matl/kilo.pdf (Baylisa
presentation), where they have booted about 1K diskless systems. I am
not sure of what horsepower is needed on the backend to support such.
However, you can imagine that you can switch over to updated versions
of OS's by simply rebooting your nodes and pointing them to an updated
image.


>
> 1) How to handle network traffic in cases of power outages 1000+
> servers going offline. I could possible see a scenario where n number
> of hosts all try to pull down a new image at the same time which could
> be any where from 50 ~ 200MB of data over the network. I would have to
> possibly implement some kind of QoS for really important servers or
> make sure I have enough Repo's close enough to minimize network
> saturation.

Perhaps leave the systems off at power up and have a scheduled power
on (WOL) based on  groups of machines in decreasing order of
importance....


> 3) Local changes to the OS, No matter how much we try to stream line
> our OS's we usually get a few admins/developers who must have
> particular changes in the OS that does not confirm to our base build.
> ie changes to something in /etc/* or /etc/{sysctl.conf,system}  if
> each server will pull down a new image at boot/reboot I need to figure
> out how to merge back any local specific changes that were on the host
> previously.

This can (allegedly) be solved by the solution above.., just maintain
a couple of additional images.

>
> 4) Boot time is another concern it currently takes most servers at
> least our 1 to 8 way boxes that makes up about 85% of the servers on
> the network  1~4 min to boot up from a complete power down.
> Theoretically would it be possible to pull down a new image, merge any
> local changes the servers had before and mount all needed file systems
> in under 5min?

Again, this can (allegedly) be solved by the solution above..as you
would have only booting time(s) to contend with

Disclaimer:  I do not work for Netapp, nor am I an expert with it..

-Cyrus