The subject of cloud costs keeps popping up in IT circles, most likely the result of more than two years’ worth of experience in shifting enterprise workloads off of traditional data center infrastructure. Increasingly, though, it seems that the cloud is not always the best choice for the pocketbook, particularly when long-term, scale-out architectures are needed.
I touched on this last month when I discussed a number of new analyses that claim internal enterprise resources can be delivered quite efficiently and at broad scale provided they are housed on the same virtual, federated infrastructure that powers most cloud services. Rob Enderle, for example, pointed out that private clouds can come in at half the cost of leading public services depending on the type of workload and the amount of data involved. A key factor in this disparity turns out to be rogue cloud deployments, which can often lead to redundancy and data duplication.
The cloud, then, represents the same buy-or-lease choice that confronts most car buyers. The lease generally puts you behind the wheel for less money upfront, but then leaves you with little or nothing at the end of the term. UK cloud provider SkyScape uses the same kind of logic to convince data users that the cloud is cheaper: roughly £36,000 for a six-week instance vs. £92000+ for dedicated physical infrastructure. This is fine if you need only temporary infrastructure for a short-term project, but it leaves out the obvious fact that dedicated infrastructure can be amortized over three or even five years and therefore becomes the cheaper option for organizations that have a continuous need for dedicated infrastructure.
Also, cloud costs are going up, either through direct rate increases or increasingly complicated license and service agreements. Microsoft, for example, recently introduced a new Server and Cloud Enrollment (SCE) program that provides new license discounts on the order of 15 percent. The problem is, earlier programs that SCE replaces offered discounts of 20 to 40 percent depending on service levels, upgrades and other factors. According to licensing guru Paul DeGroot, most customers should see licensing costs increase by 25 to 30 percent.
Still, it’s hard to argue against long-term cloud costs when services like storage are dropping in cost to as little as a penny per GB. Amazon and Microsoft, in fact, are in a race to the bottom, with both firms introducing double-digit percentage reductions at a time when data loads are going through the roof. Increasingly, though, the trade press is highlighting stories of start-ups like Moz, which launched on cloud infrastructure only to build private infrastructure when data volumes got too large. The company now employs a mix of internal and external resources, using the cloud primarily for stateless processing and related functions.
The moral of these and other such stories is that both internal, physical infrastructure and the cloud can be highly functional, cost-effective solutions for the right data sets and applications. To be sure, there will be organizations that embrace the cloud fully, just as others will resist it to the very end, but overall it seems the vast number of organizations will embrace a combination of the two. And as traditional plant becomes more “cloudy” as a result of normal refresh and upgrade cycles, hybrid cloud infrastructure is likely to become the dominant theme across the IT industry.
And the cost factor in all this? Less for overall infrastructure, but probably more for the automation and orchestration needed to determine the best way to utilize this diverse set of resources for any given workload.