Cloud Front Office

Love in the cloud is cheap!

Pretty cool use of the cloud by eHarmony. Your next important someone maybe found through cloud processing. 

Online dating site eHarmony is also making good use of Elastic MapReduce, processing tens of gigabytes of data representing hundreds of millions of users, each with several hundred attributes to be matched. According to an article on SearchCloudComputing.com, they are doing this work for $1,200 per month, a considerable savings from the $5,000 per month that they estimated it would cost them to do it internally.

Amazon Elastic MapReduce now available in Europe | Cloudiquity.

So the cloud makes love cheap? Or love is in the cloud?

I'll stop now.

Understanding cloud computing costs in the enterprise « Yet Another (ex-)VC Blog

Understanding cloud computing costs in the enterprise « Yet Another (ex-)VC Blog.

Although cloud computing provides financial benefits like reduction of CAPEX and the ability to pay-as-you-go, organizations will still need a reasonable amount of granularity in the reporting of cloud usage and the ability to map that usage into a financial chargeback model that makes sense. Knowing which applications and departments are driving IT expenses is critical now, and will continue to be critical as cloud computing goes mainstream in the enterprise. Therefore, any cloud chargeback solution should integrate with the chargeback framework that the company uses to manage their physical assets.

What's the difference between private cloud and public clouds?

There's a nice discussion going on over at Cloud Use Cases about the difference between private and public clouds. Below is my comment.

I tend to think of cloud computing as a business model underpinned by virtualization. Just as the virtualization layer abstracts the underlying hardware, there's a front-layer that abstracts the cost, sla, workload definition from the underlying computing infrastructure. 

We,and many startups, are using EC2 because the low, standard, predictable and variable cost with elastic provisioning of resources in minutes. And this is all accessed on a self-service basis with full control of the life cycle.  This change is hugely revolutionary vs what we faced before with dedicated hosting. 

Now when I hear customers talking about building a private cloud, generally they mean the ability to deliver all or a subset of the "standard, predictable variable, elastic, self-service, etc" value.  So where's the difference between private cloud and public cloud? It is wholly dependent on your role.  If you are building the technology,the technologies and issues are very similar.  If you are a consumer, you'd like to treat them both the same, 

But if you are the owner or operator, the private cloud is very challenging. You need to think of a couple of things, One, what are the standard environments and workload variations I will allow? How do I condense everything we do into a predictable unit cost?  And, how do I provide elasticity?  There are many other challenges as well, but those three are biggies for your typical IT geek.

Busting the nine myths of cloud computing | Cloud Computing - InfoWorld

Busting the nine myths of cloud computing | Cloud Computing - InfoWorld.

"Currently, it takes a lot of footwork, and a lot of manual reconfiguration, and lots of engineering effort" to move applications among public and private clouds, says Staten. And even then, "we're still in the 'I hope it works' phase."

Seamless integration is easier if customers are running the same platforms in both the public and private clouds, he says, but for the typical, more complex environments standards efforts such as the Open Virtualization Format are still "very basic" attempts to ease interoperability.

Towards a new definition of workload #cloud

This article got me thinking about what we mean when we talk about workload. 

Private enterprises have adopted some kinds of virtualization, but UCS allows them to converge server, computer and network virtualization into a single layer. That, in turn, boosts manageability of the application workload. “The ability to know that your workload can choose from a very large pool of resources to run on is a very critical element,” said Urquhart. VMware’s Herrod agreed that this kind of deeper level of virtualization, which further separates the logical view of a process from its physical implementation, makes it easier for companies to innovate. Urquhart stressed, though, that UCS isn’t just about virtualization — it is about a way to leverage fixed components in a very dynamic way.

Dynamic management of workloads, ultimately, was another reason many of the panelists agreed private clouds have emerged, and systems that enable companies to implement workloads based on actual needs will be important. Hollis of EMC said companies need to create use cases for each workload — do it all at the server layer, or all in the network, or in some cases leveraging a homogenous storage array. “You ought to be having a discussion about the functionality and less about where in the stack it happens,” he said.

We need to think of what's the right service definition abstraction that can drive the policy layer which defines how to provision and change a workload. 

Pragmatically, this means that we need to bring in a marketing layer to our offerings so that users know, in their language what IT provides.  For example, a "document collaboration site for vendos"  This then translates into, "Linux, Apache, Document Management, SQL, etc", which gets further decomposed into configuration information like "need to write new rule for firewall, need to provision VPN."

These three layers need a sort of service lifecycle manager and service catalog to keep them coherent.  

More another time.

No Catalog, No Cloud

Cloud computing is an operational model built on top of virtualization.  This operational models depends on:
1) Standardization of configurations
2) Clear offerings for environments
3) Clear resource and workload definitions
4) A self-service configuration system for customers
5) Lifecycle management by customers (what's on, what needs to go off, change, etc)
6) Basic resource accounting and charge back
7) Service level reporting.

When you put it all together, you have defined a service catalog.  So if you are starting with virtualization on your way to offer an internal or private cloud... remember:  No catalog, No cloud.