Linux

I think it's just nostalgia at this point. Kind of like all those
people who still want to petition IBM to release the source code to
OS/2's WorkPlace Shell so that it can become The Next Great Desktop
(tm).

Well they can't, or at least shouldn't. OS/2 was spun off/sold to another business which is still selling & supporting the product. The code might no longer be IBM's to release.

migrated our last remaining CentOS instance to Ubuntu today. Kind of sad because it's a superior OS to Ubuntu in most respects that matter, it's just not nearly as well-packaged To The Cloud(tm)(barf).


I expect this will change when CentOS 6 gets nice supported PV-GRUB images built, but that doesn't help us **now**.

That's been my primary complaint in recent times ... there's so much in the default build which you have to turn off, particularly if you're installing to a virtual machine. pcscd? Seriously? Does anyone still use those cards?
And then of course they pre-install the "make everything broken by default"
package, also known as SElinux.

I'm sure that most shops will build and use a clonable image for deployment to Teh Cloud (private or public) and we do that too, but it's still a hassle.

I do like the Ubuntu/Debian practice of not enabling much of anything in the default server installation. Although not even enabling the SSH server might be a bit *too* conservative -- they should at least ask about that during the setup.

I'm sure that most shops will build and use a clonable image for
deployment to Teh Cloud (private or public) and we do that too, but
it's still a hassle.

Well you shouldn't have to do this. Distributors should package a sensible image for you, with all that physical-machine stuff turned off, ssh turned on and just about nothing else.

And our management tools provider (RightScale) does this. The images they provide are everything you want in most respects, and you can get started quickly with building install scripts to layer your own stuff on top of the image. All well and good.

The only problem is this one little niggling pesky detail - the images aren't PV-GRUB based, so you can't upgrade the kernel without rebuilding the image. If the images were just PV-GRUB based, we'd be able to do "yum update kernel-xen && reboot", or incantations to that effect.
Since I refuse to fall into the trap of maintaining our own images, Ubuntu it is. I might have stuck with CentOS and bit the bullet and built an image, if 99% of our servers weren't on Ubuntu already before I started working here...

waitwaitwhat ... you're running Xen with the guests in para mode? Is that still something that's done? I thought the generally accepted practice nowadays was to either use the stock kernel with paravirtualized disk and network, or to run containers.

PVM guests account for the vast majority of Amazon EC2 deployments. Amazon requires HVM guests on their Cluster Compute and Cluster GPU instance types. All other instance types have historically used PVM-based images, and that might be the only type still supported.

That said, it's reportedly feasible (except that nobody I know of has worked out all the details yet) to boot other OS's, such as FreeBSD, using PV-GRUB; I assume this would require FreeBSD support paravirt_ops.

FreeBSD on Amazon, limited support: http://www.daemonology.net/freebsd-on-ec2/

How do I tell whether my EC2 guest is a PVM? uname doesn't seem to offer any clues.

Every AMI is flagged with a virt type, PVM or HVM. Also, if you launched with any AKI other than one of the PV-GRUB AKI's, you're definitely using PVM. If you're using PV-GRUB, it depends on how your kernel was built. pv-grub should be able to chain into any kernel that supports either the Xen 3.0.2 interface or paravirt_ops.

I suspect that the only
EC2 instance types that support HVM are:

* MS Windows instance types. (which are always more expensive than the corresponding Linux instance types.)
* Cluster Compute and Cluster GPU instance types.

Oh and of course if any program is not tracked properly by upstart,
the bug is declared to exist within *that program*, and it must find a

way to issue syscalls that upstart can grok.

you're starting to sound like me. :-) Good man.

relogin again. Of course, progress is good and variety too, but why
do they need to become ever shittier?

You're starting to sound like me too! :-)

So, I've been putting-off updating my Ubuntu10.10 systems to 11.04
because last time I tried, it screwed up xorg and nvidia horribly. I
don't want to nuke my system and do a fresh install of 11.04, nor do
I want to lose my gfx drivers and have to do back-handed ways to get
them back.

Anyone have advice?

I'm in exactly the same boat. Every time I upgrade I waste hours fixing things.
So now what I do is this: every time I start a shell and says something about 'natty' (took me a while to figure out what that even was)... This is what I do: I IGNORE IT.

I don't really like the Unity desktop, so I logged out and chose the
"Ubuntu Classic" session at the bottom during login.

I've heard enough things about gnome 3 (I gather that's what unity is?) that I fear upgrading. I'm tired of asking to have my machine broken and made worse.

I wonder if there's anything it would be good for today.

replacing ios and android sounds like a good start. :-)

Jul 28 2011 11:21am from LoanShark @uncnsrd

I suspect that the only
EC2 instance types that support HVM are:

* MS Windows instance types. (which are always more expensive than the

corresponding Linux instance types.)
* Cluster Compute and Cluster GPU instance types.

Wow, this doesn't even sound like linux anymore. I'm so 19th centurry.

You're so two thousand and late.

Jul 29 2011 12:00am from LoanShark @uncnsrd

You're so two thousand and late.

OHHH!!! I know that song!!!