As we approach the end of 2009, the hands-down winner of the hot IT trends award continues to be cloud computing. Yet the reality is that for many, cloud deployment strategies remain somewhat ethereal — part theoretical, part vendor/marketing hype, and part vaguely defined future model.
At the same time, the intensity of interest in the cloud approach has been almost unprecedented. The promise of the cloud has struck a deep chord that resonates strongly, making it a phenomenon that IT organizations simply cannot ignore — or do so at their peril.
In one sense, the profound interest in the cloud model can be viewed as an indicator of a rather profound level of dissatisfaction with IT as it is currently practiced. Often viewed as unresponsive or at least non-empathetic, IT is too often perceived as an inhibitor to progress — and a very expensive one at that. Along comes cloud computing, promising unimpeded flexibility, instantaneous provisioning, and unlimited capacity, with no upfront investment and pay as you go pricing. At 50,000 feet, at least, what’s not to like?
Challenges come at lower altitudes, and much has already been written about hurdles related to performance, security, availability and other attributes. Clearly, cloud computing is still maturing: In its “Hype Cycle for Cloud Computing, 2009,” Gartner indicates that the vast majority of cloud computing technologies are “more than two years from mainstream adoption.”
As cloud options become more viable, they will begin to undergo more detailed economic scrutiny, but it’s worth examining the state of cloud economics today and how this model is likely to evolve.
Earlier this year, McKinsey & Company published a widely reported and much debated discussion document entitled “Clearing the air on cloud computing” (registration required), which included an analysis suggesting that the costs for many of Amazon’s EC2 services rather quickly exceeded those of a traditional IT infrastructure when scaled to mid- to large-sized environments.
Since then, there have been several other efforts to evaluate “traditional IT build” versus “cloud buy” using Amazon’s pricing as comparison points. It’s interesting to note that Amazon is consistently the benchmark used in these comparisons presumably because it is widely regarded as a cloud infrastructure leader, but especially because its prices are incredibly easy to obtain. This is, in itself, a big deal.
How many IT organizations can provide the true costs for compute and storage services so readily? Regardless of the actual level of service, the cloud providers can at least quickly tell you what you get from their service and how much it will cost.
Invariably, these comparisons have demonstrated that it is often still technically cheaper to build server infrastructure configurations than to buy them, but with some caveats. One fundamental consideration is what assumptions are made concerning utilization. If systems are in use 24x7x365, then the build scenario looks more attractive.
Since Amazon bills by the hour, if a server is not fully utilized, then at some point the cost balance tips to Amazon’s favor. Low utilization of servers is far from unlikely — in fact, it is the high probability of low utilization that underlies the success of server virtualization. Our virtualization practice regularly identifies large quantities of servers with CPU utilization rates of less than 10 percent!
Another major factor in cloud economics is nontechnical, relating to corporate finance objectives and practices. Depending on the organization, having a balance sheet that is heavy with capital assets may be viewed as undesirable, and IT spending often impacts this significantly.
In organizations for which the cost of capital is high or that have other reasons for wishing to minimize capital purchases, the cloud approach, like other hosted options, affords an opportunity for cap-ex avoidance.
Unlike other options that typically involve time commitments and other contractual obligations, cloud services can be expanded or contracted on a much more granular basis. For some, this degree of flexibility alone may be enough to tip the scale in favor of the cloud.
The avoidance of cap-ex is clearly an economic attraction for public cloud services, but what about private cloud infrastructures, with IT organizations building and operating their own cloud environments? Here the picture gets a bit cloudier. (Sorry!)
An internal cloud certainly requires infrastructure, which means buying (or leasing) gear to support demand. The key opportunity — and challenge — is achieving greater efficiency through improved utilization. However, this requires that IT adopt a model — and mentality — similar to that of a service provider, including the following:
- effective capacity planning to avoid buying too far ahead of the curve;
- standardized service offerings to simplify provisioning, maximize utilization, and minimize the inefficiencies of specialized, or “one-off,” builds; and
- a cost-management capability that offers simplicity at the customer-facing request and provisioning stage, but is able to manage the back-end complexities of incorporating and allocating the various cost elements.
This is a tall order. Furthermore, all but the largest organizations are likely to be at somewhat of a disadvantage relative to public cloud providers when it comes to economies of scale.
For these reasons, external hosting providers that can offer services that act as an extension to the internal private cloud are likely to play a growing role in IT cloud plans. While not necessarily offering the instant on/off option of public providers, they do afford companies some flexibility to, for example, expand or contract their infrastructure based on cyclical needs, mitigating the requirement to purchase infrastructure capacity whose primary purpose is to accommodate infrequent peak loads.
The Cost of Five 9s
A crucial factor to consider when evaluating any cloud option is to understand exactly what you are getting for a given price and how well it aligns to what you need. Most public cloud infrastructure vendors currently position their services in the range of three 9s or so (99.9 percent to 99.95 percent) of availability. For most enterprises, availability expectations for production platforms are more likely to land in the five-9s realm, which would make the cloud a non-starter in these situations. Can this level of availability be addressed by the cloud? If so, at what cost?
A recent article detailing the Pentagon’s Rapid Access Computing Environment (RACE) cloud services provides a somewhat startling answer. The cost of production servers offering 99.999 percent can be had at the seemingly not-so-bargain price of US$1,200 a month! Now, to be fair, the article does not provide details about what specifically is included. I would expect a system offering such a level of availability to be highly redundant, and considering the targeted user base, incorporate a high degree of security as well.
Determining whether this price is reasonable or excessive requires a specific understanding of what it would cost to deliver a comparable set of services in a traditional environment — price without context is meaningless. So, ultimately, the key to evaluating cloud economic viability is an understanding of both technical requirements and usage patterns. The ability — or lack thereof — of an IT organization to articulate these details will go a long way toward determining the appropriateness of a cloud solution.
Cloud evolution is advancing rapidly, and pricing is somewhat of a moving target (mostly downward) as technologies mature and service providers refine their models and respond to competition. The message for IT organizations should be clear: The cloud concept — whether formally embraced or not — is changing the IT conversation, and those organizations hoping to thrive must be prepared to participate in the discussion.
James Damoulakis is CTO at GlassHouse Technologies.