[wplug-internet] PaaS, instead of Hosting? -- New Linode upgrade available

[Bryan J Smith]

On Mon, Apr 28, 2014 at 9:13 AM, Justin Smith <justin at adminix.net> wrote:
> I wouldn't say that WPLUG is necessarily looking forward to Docker.

Understood.  But Linux Containers (LXC) are what most distributions
have been developing towards supporting for some time now.  The
libvirt abstraction like a VM, but at a greatly reduced footprint,
make them ideal.  Of course, security becomes on a concern with LXC,
as everyone is actually sharing the same OS.  And some have their
differences (e.g., Red Hat uses SELinux).

> As a rule, I tend to be more interested in upgrading to new technology than
> others in the group.

Well, Enterprise Linux distros tend to be more trailing edge, sans
select technologies which are sometimes backported immediately or even
a core component maintained alongside Upstream.

> Pat, I know, feels that systemd is a solution in search of a problem,

Some would feel the opposite as the lack of a general system manager
has been holding back Linux while other UNIX flavors have already
adopted something similar.  Systemd just finally ties in various,
existing Linux subsystems that were not managed well for init, much
more run-time, including and dynamic resource and service discovery
and dependencies (instead of hardcoded ones).

I look to the Debian Steering Committee.

For those that didn't follow it, major Debian users (e.g., thousands
of systems) were calling for it, hence why it found support.  CoreOS
and a lot of cloud generation platforms absolutely require something
like it.  It became unavoidable, along with its *d solutions.  Red Hat
shipped Upstart in RHEL6, even if it wasn't completely utilized for
all init processes, but many customers were still using monit and
other solutions instead.

> and most of the others are probably content to leave MySQL alone for
> the time being because they don't dislike Oracle as much as I do.
> I only started working as a system administrator in December, so some
> enterprise-level virtualization technology is new to me. Before you brought
> this up, I had never bothered to look into OpenShift even though I'd heard
> the name mentioned before. Not sure if anyone else in the group has worked
> with this.

OpenShift solved an immediate need for Red Hat's customers, especially
Middleware (JBoss).  It provided a way to instantiate multiple stacks
on the same host OS, but contain them via cgroups, SELinux, etc...  As
a natural byproduct, other languages and different database stores
could also be provided.  RHSCL came out and allows different, newer
upstream version languages and database releases, that Red Hat will
maintain backports on for several years.

OpenShift "just works" today and doesn't require a lot of integration
or development.  Cartridges are easy to create and deploy, hence it's
popularity with a growing number of Middleware customers, and for even
non-Java solutions.  LXC support, with libvirt abstraction, was always
planned for RHEL7 and an eventual RHEL6 backport.  Now Docker is
pretty much the direction everyone wants to go, including Red Hat.
RHEL7 will have it, and there will even be a special release --
"Atomic" -- that is a very minimal host for containers.  There will be
more limited Docker solution backported to RHEL6.

OpenShift shouldn't be confused with OpenStack, even thought there is
some overlap, OpenStack is a massive undertaking designed to address a
much broader solution set.  OpenStack is designed to do LXC as well as
VMs, although if you've been exposed to OpenStack ... I like to call
it "OpenSnack."  I.e., it's not quite a full meal yet.  Eventually
anyone with an OpenStack Platform will be able to manage just about
any Cloud generation app.

The main thing that assumption that is totally wrong about OpenStack
is that it's a VMware replacement.  Apples-to-Oranges.  It's a cloud
generation platform for cloud apps.  Although one can use OpenStack
for traditional virtualization, that's wasting a lot of OpenStack's
capabilities.  Instead, you'd use a VMware farm as a OpenStack Nova
Compute hosts, maybe some of the other, farm capabilities, as long as
they map into OpenStack components -- e.g., Storage for Cinder,
Glance, Swift, etc...  The open source equivalent of VMware vSphere is
the oVirt et al. abstraction layers (some also used in OpenStack)
which is best known as Red Hat Enterprise Virtualization (RHEV).  RHEV
too can provide OpenStack Nova Compute hosts.  Etc...  Traditional,
managed virtualization don't "compete" with OpenStack, and OpenStack
doesn't replace them, as OpenStack has much broader goals overall.

> I don't think RAM usage would be a problem. Free -m, adjusted for buffer
> usage, says that we're not even using half a gig of RAM. We'd have to use
> something else for storage space, though. We need a place to store
> membership rosters, organizational documents, and such, and $1/GB/month for
> OpenShift storage is just ridiculous.

Well, 1GB is free, actually 3GB -- 1GB/gear, up to 3 small gears for free.

And that's just OpenStack Online, a free PaaS.  OpenStack is a product
from Red Hat that is available to corporations for their own hosting,
internally or externally.

> It's an interesting possibility. I'll play around with OpenShift if I have
> time this week.

I only mentioned it because OpenStack Online could be used for free
prototyping of a new WPLUG site with various language choices and
database versions that will be maintained for years -- thanx to
RHEL+RHSCL.

And maybe some portions of the site could just be hosted there, where
the data fits in 3GB total.

-- bjs