[arch-general] Archlinux hypervisor

[C Anthony Risinger]

On Mon, Nov 22, 2010 at 5:54 PM, Matthew Monaco <dgbaley27 at verizon.net> wrote:
> On 11/22/2010 11:41 AM, C Anthony Risinger wrote:
>>
>> fellow patrons,
>>
>> i am moving my company's in house platforms off of a mix of deprecated
>> soft virtualization solutions like openvz and bare metal installs -- i
>> have decided on using libvirt + kvm (in conjunction with cgroups for
>> qemu and possibly other uses).
>>
>> i'm trying to use Arch for the host/hypervisor setups, for a couple
>> reasons:
>>
>> ) very easy to roll custom initramfs builds
>> ) shooting for a<  128MB 100% in-RAM diskless image (RHEV-H or ESXi)
>> ) host is changed infrequently, and all-at-once (no need for
>> ongoing/stable repo support)
>> ) i'm very familiar with Arch's initramfs setup, and with Arch in general
>>
>> the other possibility i'm considering is Gentoo, but it's been quite
>> some time since i've used it :-/.  neither arch nor gentoo will be
>> used for guests (likely debian/ubuntu here).
>>
>> does anyone have any experience/insight here?  the comparable
>> technology for RHEV-H for redhat, or ESXi for vmware.  i am looking
>> into the `archboot` setup for inspiration on a custom initramfs build
>> system.
>>
>> primarily i'm just look for any pointers/information anyone may find
>> useful, not only on the image itself but also management and
>> configuration techniques.
>>
>> thanks,
>>
>> C Anthony
>>
>
> How important is the stability of your system? Will you be able to build
> custom versions of packages when security updates are released so that you
> don't have to start upgrading everything?

well since these are the host images, i pretty much want them to be as
rock solid as possible.  there is not much need for packages/updating;
the host systems will be network booted (or booted from a small
flash... maaaaaybe CDROM as last resort), and are a completely self
contained image -- kernel + initramfs.

everything inside the image will be hand picked and stripped down --
no docs/etc. only the necessary kernel modules, only the bare minimum
services needs to run VMs, diagnostics, and metrics.  kernel will be
hand tuned.

the idea is that with a couple host machines in place, and a
distributed block layer (undecided... LVM + iSCSI or something
else...), VMs can be migrated off a host, and the host is simply
rebooted to run a new, updated image.  VMs can then be moved back
without downtime.

i'm not too concerned with host security (guests are a diff story...);
the host images are totally inaccessible except for SSH, and via a
built in host key.  we might allow TCP + TLS connections, but
unlikely.

so yes, we will be doing many custom packages and builds; i'm trying
to see if anyone has done custom initramfs work like this, beyond the
archboot tool (which is great :-).

C Anthony