Wednesday, November 17, 2010

Cloudy in the Private Public and Hosted worlds

Today there was a thread in the forums that brought up the new buzz word: “Private Cloud”

There is currently a lot of hype, questions, and confusion about this term – it is not a new term by the way – since it was used at TechEd 2010 in Berlin.

The original poster appears just as confused as the majority of folks as to what this term might mean.  And in the industry so far this seems to be the current state of things.  The question:  “What is a cloud?” 

My take on “cloud” is that it is a pool of compute resources that works in some way and it is in some place.  That seems to be the cross vendor understanding of cloud.

Today there are a few “cloudy” terms floating around.  Private, public, and hosted.

Private Cloud is a cloud that your (as an enterprise) own and run.  It is your traditional datacenter.  But set-up and managed in a way where it does not matter where the application runs as long as the application runs.

Public Cloud is a cloud that is “out there” and runs at a facility owned by someone other than you.  And the key is that you purchase compute resources on that cloud.  Azure is a really good example of this.  Amazon EC2 as well.

Hosted Cloud is a cloud that is “out there” and runs at a facility owned by someone other than you as well.  They key here is that it is provisioned and billed more like a co-location facility.  You have “machines” that you purchase – not highly generic capacity.  This is more like the Rackspace or upcoming Azure VM Role models.


The most interesting part of the entire thread that caused me to begin this post in the first place is that HPC was mentioned by the original poster.  You know, you could think of HPC as a private cloud.  Just as a good ‘ole mainframe could be a private cloud.

Here is how I responded to the thread – simply to broaden the thinking beyond a simple hypervisor:

I wanted to jump in to this thread as there are two distinctly different technologies being discussed. And since the OP is coming from the education industry both or neither could apply.

The term "Private Cloud" is simply a set of processing resources that you own and therefore control. It is not remote. That is all that we really know today. Beyond that, it is not clear if this is a combination of technologies, a SKU bundle, a pre-installed hardware package, etc.

HPC, Azure Appliance, and Hyper-V + SCVMM are all options that would give you something that you could refer to as a private cloud.

HPC is a platform designed for distributed processing - many individual nodes that provide CPU and contribute back to the primary job. This can be thought of as very mainframe-ish. I over-simply the concept and think of it as the SETI@Home or Folding@Home model - as it is a large compute. This requires custom written applications.

Azure is a platform - that is making breaks in the education realm right now - but it is not private. It is a pool of processing resources that can be scaled out and back on demand. It is HPC-like in concept, but Hyper-V-like in large scale implementation. MSFT owns and manages the hardware, you own the application - you purchase capacity during a slice of time on the Azure platform. This requires custom written applications.

Azure Appliance - this is the cloud version of Azure, but you own it, it is your datacenter, you house the physical hardware - all the other Azure rules apply. You simply don't pay MSFT for time as you own the hardware.

Hyper-V + SCVMM - this is the current model of using virtual machines to run enterprise applications. This allows you to run off-the-shelf applications installed on to operating systems - just as you would with physical hardware. Hyper-V is the hypervisor, SCVMM is the management stack for the virtual machines. Adding the VMM SSP 2.0 in front of SCVMM provides additional functionality.

Each one of these fits in different places and for different reasons. And they could all be referred to as "cloud" in some way or another. As it is very unclear what "cloud" really means beyond being a pool of compute resources.

In the end, it really comes down to what you require. Large complex calculations would gravitate to HPC or Azure. More traditional workloads would gravitate toward Hyper-V + SCVMM.

No comments: