Rent vs. Buy? The Cloud Conundrum

Over the long-run, is cloud computing a waste of money? Some startups and other “asset lite” businesses seem to think so. However, cloud computing for specific use cases, makes a lot of sense—even over the long haul.

Courtesy of Flickr. By IntelFreePress
Courtesy of Flickr. By IntelFreePress

A Wired Magazine article emphasizes how some Silicon Valley startups are migrating from public clouds to on-premises deployments. Yes, read that again. Cash poor startups are saying “no” to the public cloud.

On the whole this trend seems counter intuitive. That’s because it’s easy to see how capital disadvantaged startups would be enchanted with public cloud computing: little to no startup costs, no IT equipment to buy, no data centers to build, and no software licensing costs. Thus for startups, public cloud computing makes sense for all sorts of applications and it’s easy to see why entrepreneurs would start—and then stick with public clouds for the foreseeable future.

However, after an initial “kick the tires” experience, various venture capital sponsored firms are migrating away from public clouds.

The Wired article cites how some start-ups are leaving the public cloud for their own “fleet of good old fashioned computers they could actually put their hands on.”  That’s because, over the long run, it’s generally more expensive to rent vs. buy computer resources. The article mentions how one tech start up “did the math” and came up with internal annual costs of $120K for the servers they needed, vs. $320K in public cloud costs.

For another data point, Forbes contributor Gene Marks cites how six of his clients analyzed the costs of public cloud vs. an on-premises installation, monitored and managed by a company’s own IT professionals. The conclusion? Overall, it was “just too expensive” for these companies to operate their workloads in the public cloud as opposed to capitalizing new servers and operating them on a monthly basis.

Now to be fair, we need to make sure we’re comparing apples to apples. For an on-premises installation, hardware server costs may be significantly less over the long run, but it’s also important to include costs such as power, floor space, cooling, and employee costs of monitoring, maintaining and upgrading equipment and software. In addition there are sometimes “hidden” costs of employees spending cycles procuring IT equipment, efforts for capacity sizing, and hassles of going through endless internal capitalization loops with the Finance group.

Thus, cloud computing still makes a lot of financial sense, especially when capacity planning cycles aren’t linear, when there is need for “burst capacity”, or even when there is unplanned demand (as there often is with fickle customers). And don’t forget about use cases such as test and development, proof of concept, data laboratory environments and disaster recovery.

Another consideration is resource utilization.  As I have stated before, if you plan on using IT resources for a brief period of time, cloud computing makes a lot of sense. Conversely, if you plan on operating IT resources at 90-100% utilization levels, on a continual and annual basis, it probably makes sense to acquire and capitalize IT assets instead of choosing “pay per use” cloud computing models.

Ultimately, the cloud rent vs. buy decision comes down to more than just the price of servers. Enterprises should be careful to understand their use cases for cloud vs. on premises IT. In addition, watch for hidden costs in your TCO calculation that underestimate how much time and effort it really takes to get an IT environment up, running and performing.

Private Clouds Are Here to Stay—Especially for Data Warehousing

Some cloud experts are proclaiming private clouds “are false clouds”, or that the term was conveniently conjured to support vendor solutions. There are other analysts willing to hedge their bets by proclaiming that private clouds are a good solution for the next 3-5 years until public clouds mature.  I don’t believe it. Private clouds are here to stay (especially for data warehousing)—let me tell you why.

For starters, let’s define public vs. private cloud computing.  NIST and others do a pretty good job of defining public clouds and their attributes.  They are remote computing services that are typically elastic, scalable, use internet technologies, self-service, metered by use and more.  Private clouds, on the other hand are proprietary and typically behind the corporate firewall. And they frequently share most of the characteristics of public clouds.

However, there is one significant difference between the two cloud delivery models –public clouds are usually multi-tenant (i.e. shared with other entities/corporations/enterprises). Private clouds are typically dedicated to a single enterprise – i.e. not shared with other firms. I realize the above definitions are not accepted by all cloud experts, but they’re common enough to set a foundation for the rest of the discussion.

With the definition that private clouds equate to a dedicated environment for a single or common enterprise, it’s easy to see why they’ll stick around—especially for data warehousing workloads.

First, there’s the issue of security. No matter how “locked down” or secure a public cloud environment is said to be, there’s always going to be an issue of trust that will need to be overcome by contracts and/or SLAs (and possibly penalties for breaches).  Enterprises will have to trust that their data is safe and secure—especially if they plan on putting their most sensitive data (e.g. HR, financial, portfolio positions, healthcare and more) in the public cloud.

Second, there’s an issue of performance for analytics.  Data warehousing requirements such as high availability, mixed workload management, near real-time data loads and complex query execution are not easily managed or deployed using public cloud computing models. By contrast, private clouds for data warehousing offer higher performance and predictable service levels expected by today’s business users. There are myriad other reasons why public clouds aren’t ideal for data warehousing workloads and analyst Mark Madsen does a great job of explaining them in this whitepaper.

Third, in the multi-tenant environment of public cloud computing, there is increasing complexity which will lead to more cloud breakdowns. In a public cloud environment there are lots of moving pieces and parts interacting with each other (not necessarily in a linear fashion) within any given timeframe. These environments can be complex and tightly coupled where failures in one area easily cascade to others. For data warehousing customers with high availability requirements public clouds have a long way to go.  And the almost monthly “cloud breakdown” stories blasted throughout the internet aren’t helping their cause.

Finally, there’s the issue of control. Corporate IT shops are mostly accustomed to having control over their own IT environments. In terms of flexibly outsourcing some IT capabilities (which is what public cloud computing really is), IT is effectively giving up some/all control over their hardware and possibly software.  When there are issues and/or failures, IT is relegated to opening up a trouble ticket and waiting for a third party provider to remedy the situation (usually within a predefined SLA).  In times of harmony and moderation, this approach is all well and good. But when the inevitable hiccup or breakdown happens, it’s a helpless feeling to be at the mercy of another provider.

When embarking on a public cloud computing endeavor, a company or enterprise is effectively tying their fate to another provider for specific IT functions and/or processes.   Key questions to consider are:

  • How much performance do I need?
  • What data do I trust in the cloud?
  • How much control am I willing to give up?
  • How much risk am I willing to accept?
  • Do I trust this provider?

There are many reasons why moving workloads to the public cloud makes sense, and in fact your end-state will likely be a combination of public and private clouds.  But you’ll only want to consider public cloud after you carefully think about the above questions.

And inevitably, once answers to these questions are known, you’ll also conclude private clouds are here to stay.

 

Private Clouds Better for Data Warehousing than Public Clouds

Cloud solutions are useful when additional computing power is needed. And cloud capabilities are relatively easy to procure because customers can sign up with a credit card—via an online portal—and start using services within minutes. This is the public cloud delivery model of Amazon and Google, among others.

Public clouds, however, can have a “not-so-silver lining,” with documented concerns over security, privacy, availability, data loss and latency issues. Organizations wishing to mitigate risks associated with storing and analyzing sensitive data in public clouds are increasingly turning to private clouds.

Public cloud solutions generally satisfy user expectations for applications like sales force automation or marketing campaign management. However, data warehousing requirements such as high availability, mixed workload management, near real-time data loads and complex query execution are not easily managed or deployed using public cloud computing models. By contrast, private clouds for data warehousing offer the higher performance, better security and predictable service levels expected by today’s business users. Read more

Is there Too Much Complexity in the Cloud?

A recent analyst report suggests public clouds are prone to failure because they are inherently complex. However, just because there are multiple interacting objects in a particular environment, this doesn’t necessarily imply complexity.

Cloud computing is all the rage for business users and technology buyers. And why not, especially because it provides a fast and flexible option for delivering information technology services. In addition,cloud computing also drives value through higher utilization of IT assets, elasticity for unplanned demand, and scalability to meet business needs today and tomorrow.

However, there are risks in the cloud, especially in the public cloud where business and news media regale with case studies of data loss, security issues, failed backups and more. Perhaps one reason that public clouds are prone to failure—and perhaps always will be—is that some analysts consider these environments to be complex and tightly coupled.  And if indeed this is the case, then IT buyers must consider that failure isn’t only possible, it’s inevitable.

Yet, first we must ask, are public clouds really complex environments?

To understand if a particular system is complex, we must understand if it has characteristics such as connected objects (nodes and links with interdependencies), multiple messages and transactions, hierarchies, and behavioral rules (instructions).

Public cloud services available from companies such as Microsoft, Google, Amazon Web Services (AWS) etc., often consist of various components such as applications (front end and backend such as billing), controllers and message passing mechanisms, hardware configurations (disk, CPU, memory), databases (relational and NoSQL), Hadoop clusters and more.  In addition there are various management options (dashboards, performance monitoring, identity and access) and these environments typically operate with multiple users, multiple tenants (compute environments shared with more than one application and/or company), and sometimes span multiple geographies.  And from a complexity standpoint we haven’t even yet discussed processes in building cloud environments much less operating them.

In summary, in a cloud environment there’s lots of moving pieces and parts interacting with each other (not necessarily in a linear fashion) within any given timeframe.

Multiple interacting agents can help define whether a particular environment is complex or not, however another key determinant is also very important—whether processes are tightly or loosely coupled. Richard Bookstaber, author of Demon of Our Own Design, writes that tightly coupled systems have components critically interdependent with little to no margin for error. “When things go wrong, (an error) propagates linked from start to finish with no emergency stop button to hit,” Bookstaber says. So a tightly coupled system is one where linkages (dependencies) are so “tight” that errors or failures cascade and eventually cause the entire system to fail.

This discussion is important from a risk management perspective for cloud computing. If we believe that data is one of the most valuable assets of a corporation and if we believe public clouds are complex environments with tightly coupled components that have little to no slack (buffers) to stop failures, then there should be a set of practices and processes set in place to manage the potential risk of data breach, theft, loss or corruption.

So what say you, should public clouds be considered “complex” environments?  Are they “high risk” systems prone to failure?