posted

0 Comments

By Steve Jin, VMware engineer

 This entry was reposted from DoubleCloud, a blog for architects
and developers on virtualization and cloud computing.


What’s Different in Cloud Computing?

If you look at the varieties of software at
different levels of the stack, you will find the lower the stack the fewer
number of choices. When you move up to middleware, you will have more choices. At
the application layer, the growth of possibilities gets exponential.

If we draw a diagram here, it’s then like
an inverted pyramid shown in following diagram
.


Pyramid

Installing an OS or Cloning from template?

With the use of virtualization in cloud
computing, the demand for the OS provisioning tools will be much less than
before. For one thing, OS plays a less important role for applications. Most of
the time it doesn’t matter what OS is there as long as there is one. It
therefore makes it possible to consolidate the OSes to several standardized
flavors.

On the other side, virtual machines are
really just a bunch of files. You can save the standardized OSes as templates
or machine images in Amazon terms. When you provision a new machine, you
include the OS as well. There is not much need to
install it by yourself.

Still, service providers may see many
different types of standard OSes even though each user just uses a handful. The
service providers therefore still need to use OS provisioning tools to accommodate
all the different requirements of their customers.

What about the extra storage space? Good
question. Overall the portion of OS images is relatively small and manageable
for the service providers who can cross leverage them for different users. Also
modern storage de-duplication software has significantly reduced actual storage
use, especially when the OSes are mostly the same (for example, Linux OSes
of different flavors and different versions).

When it comes to
the format of virtual machine template, you want to follow the
OVF spec which is a DMTF standard supported by major
players.

How About Middleware?

When it comes to middleware, things can get
tricky. There are definitely more combinations than with
operating systems. It may or may not be a good idea to pre-install the middleware inside
a machine template.

The deciding factor is really how much you use the particular middleware. More
often than not, you have a pre-defined set of middleware for a particular
purpose – for example, building Java-based enterprise Web applications. You can
install all the middleware components in a virtual machine image. It’s not only
easier to deploy, it can also test as the gold image
to standardize within an organization. It also saves time to maintain, patch, and upgrade the related
middleware to save costs and improve system security.

There are
certain cases in which you have a very specific combination of middleware.
Given the rarity of usage, you may either use configuration tools or install it
manually.  You decide by whichever is
more convenient. When you do something just once, you probably won’t bother
with learning tools like Puppet. But if you are already familiar with one of
the tools, why not use it?

Applications?

Applications
have the most varieties in terms of numbers and functionalities. In reality,
you may not want to create a virtual machine image for each application most of
the time. That’s especially true when you are in the process of developing
applications – you don’t have a stable application to build a virtual machine
image yet.

Can you wait for
the stable release of the application? In theory, it’s possible. But in
reality, it’s mostly not an option. You have to deploy it and test it in the
staging environment continuously upon daily build or even every check-in.

To facilitate
this
continuous integration and deployment, you would like to use some automation
framework as I mentioned earlier. Besides the application provisioning, you can
also automate the testing with the same framework.

Life could
really be this easy? Yes. That is the benefit of using a framework. However, no
pain no gain. The pain is to set up the framework and write/test the commands.
Once you set up the process, you cannot easily change the procedure of
installation. This is not an option for most developments especially in initial
stages, therefore you have to maintain that as well.

As I mentioned
in
my previous blog on cloud application architecture, you want
to design your applications to be as stateless as possible so that you can
standardize the application for massive deployment. If that is the case, you
can build a virtual machine with your application installed on the fly. If you
choose this approach, you should incorporate the building of a virtual machine
as part of your release process, and deploy that virtual machine instead of
pushing applications to virtual machines.
Many tools like VMware Studio can assist you to do that.

Summary

System
provisioning in cloud computing is an important aspect of system architecture,
application design and operation. As virtualization gets more and more popular
as the enabling technology for cloud computing, we will see more use of virtual
machine images/templates for system provisioning, including operating systems
and middleware. The standardized templates provide not only operational
efficiency but also high quality, pre-qualified software stacks for building
cloud applications.

Application provisioning
frameworks will still play an important role in system provisioning and other
activities such as lifecycle management, automated testing, and more. It’s not
only a tool for system operation, but also an essential tool for continuous
system integration.

VMware engineer Steve Jin is author of VMware
VI & vSphere SDK (Prentice Hall)
, founder of open source VI Java API, and is the chief blogger at DoubleCloud.org