[sdnog] help with virtualization configuration

[Patrick Okui]

Hi Sara, all

On  31-Aug-2015 22:44:15 (+0300), Nishal Goburdhan wrote:
> and you can probably get to all of this from KRT ..  ;-)
> (c/f https://twitter.com/sara_alamin_h/status/638423741165367297)


Hmm. I didn't know google code wasn't available in السودان‎
is someone able to give me access to a vpn into Sudan for testing things
like these?

In anycase, Google Code is dead
<http://www.ghacks.net/2015/03/12/google-code-is-dead/> so the project
moved to <http://git.ganeti.org> and a read only mirror on github
<https://github.com/ganeti/ganeti>. The mailing list is still on google
groups AFAIK. Is that available in Sudan?


As to your original question, Nishal's questions help steer towards a
particular set of recommendations.

And yes, VMWare will likely do what you want plus the kitchen sink. The
problem is cost. psg.com was on vmware infrastructure but we had drive
failure and for our combination of controllers and disks there was no
way to get info about which drive failed without an upgrade to the next
major version. This would've cost about the same as buying the licenses
from scratch so we didn't do that.

At the bare minimum Microsoft, FreeBSD's bhyve, Sun's Xen, Linux's KVM
will let you create VMs and with shared (and/or replicated) storage you
can migrate between VM hosts.

If you have say KVM or XEN as the hypervisor and libvirt on top of it,
you can easily do even a live migration as long as the storage is shared
so nbd or drbd. The catch with this is these solutions is they don't
view the servers as nodes in a "virtualization cluster" more like
individual virtualization hosts. Last time I used proxmox they had the
same limitations but they seem to have added some high availability
clustering <https://www.proxmox.com/en/proxmox-ve>

In my presentation I mentioned Amazon are using opensource. Their
offering is based on Xen and openstack sitting on top of that. The magic
comes from the provisioning tools they've put on top of that. There are
a few projects to give you something similar talking to the openstack
api. I've stayed away fro openstack because I find the learning curve
way too high. Too many pieces have to work just right before you can get
started creating a single VM with it.

ganeti is my current favourite for "cluster style" solutions. This is
what psg.com is running <https://wiki.rg.net/wiki/ForkLift> but gives
you a shell interface where you log in as root to the ganeti master to
issue commands.

Synnefo fills this gap by giving you a web gui for self provisioning etc
close to what you get with amazon AWS. There are still some missing
pieces by design. E.g failover incase a ganetti node collapses
(particularly the master) is not an automatic event. It has to be done
manually (or by some monitoring system) to avoid split brain issues.

The last time I did a workshop based on these concepts was early this
year at SANOG. You could take a look at
<https://nsrc.org/workshops/2015/sanog25-virtualization/wiki/Track2Agenda>
and see if you find anything useful there.

Let us know what you end up going with.

--
patrick


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 204 bytes
Desc: OpenPGP digital signature
URL: <http://lists.sdnog.sd/pipermail/sdnog/attachments/20150904/68a3a220/attachment.sig>