Three private cloud tiers have emerged, with each having an associated price range, where increasing value is justified through increasing resolution of CIO concerns:
- Cost-reduction tier – where public cloud cost effectiveness is the primary differentiator
- Overhead-reduction tier – that includes better monitoring and SLAs
- Risk-protection tier – with multiple SLAs and compliance
With managed hosted private cloud coming down 22% in the past year, it is increasingly important to show the value of bundled capability.
Public cloud has a price differentiator in the form of on-demand consumption, which isn’t as sensitive to unused capacity as most private clouds. However, the landscape is changing: Public cloud pricing models that offer a discount for commitment are often susceptible to this issue of unused capacity, which ultimately results in higher unit costs. This increasing susceptibility, plus new pay-as-you-go and build-operate-transfer models for private cloud, affects the economics of public vs. private cloud.
Private clouds owned and self-managed by enterprises can be cheaper than public cloud. The magic number to beat is about $25 per VM-month at 100% utilization. If the cost of the whole stack comes in under this number, then, even with the addition of labor to manage that private cloud, it should be cheaper than public cloud. Enterprises unable to achieve a labor efficiency of 300 VMs per engineer are unlikely to beat public cloud on price.
Partially managed clouds have good economics. If an enterprise can manage just the datacenter element of a private cloud at a ratio of least 400 VMs per engineer, that cloud may have an advantage over fully managed alternatives. We believe enterprises could easily beat this ratio.
Enterprises should determine their primary drivers for private cloud and their areas of labor strength and weakness. CIOs should be up front about these – it is the service provider’s or vendor’s job to provide a solution that matches the required capability and allays the CIO’s concerns.
Broadly, the value of one cloud compared to another can be assessed using three closely related measurements:
- Direct cost value – the reduction in unit costs achievable by using the service, compared to other options
- Labor value – the reduction in effort obtained by using a service such that overall expenses are reduced
- Revenue value – the improvement in or protection of revenue as the result of using a particular service
Public cloud possesses all of these types of value. In the Cloud Price Index: Private Cloud Edition, Managed Services report, managed private cloud can have cheaper unit costs than public cloud in a minority of circumstances.
Where revenue value is concerned, consumers that seek the lowest-priced public cloud resources through commitment are susceptible to waste, which leads to sunk costs and higher unit costs. As a result, private and public cloud are both cost sensitive to utilization.
Public cloud has the upper hand due to its ‘unlimited’ ability to scale, whereas private cloud is capped to some degree. However, hybrid and new pay-as-you-go private cloud models mean private cloud has the revenue value of being able to scale quickly.
The low cost and flexibility of public cloud has made its way on-premises in several forms. Metered hybrid stacks (such as VMware Cloud on AWS and Azure Stack) offer dedicated capacity with the ability to tap into public cloud for specialized workloads.
Hardware vendors and managed service providers (IBM, Hewlett Packard Enterprise [HPE], Dell, Nutanix, Oracle) are putting equipment on customer premises and charging based on consumption for the services running on top of it (sort of the camera/film or printer/ink model).
Meanwhile, with the build-operate-transfer model, open source vendors (Rackspace, Canonical, Mirantis, Joyent) are offering to do the heavy lifting of implementing, securing and initially operating an on-premises private cloud, with the option of returning control back to the customer for continued development and ongoing operations.
On the other hand, the easy scalability of public cloud can make it seem like the Wild West to enterprise CIOs, especially when costs spike unpredictably. Public cloud providers have responded with easy-to-understand commitment and pre-purchase discounts – think reserved instances, committed-use discounts and universal credits – trading flexibility for savings and a measure of fiscal discipline.
According to the Public Cloud Price Index: US Q3 2018 Data and Analysis, benchmark medium VM prices in the US were flat over the past year, but the average discount for a basic application became steeper than ever – 39% less than on-demand rates.
Two sets of questions are now crucial:
- How should service providers bundle capability into their offerings, and what should enterprises pay for these different service tiers?
- When should enterprises consider running their own private clouds, and how should service providers handle such situations?
These questions ultimately help us understand the labor value of managed services.
When is Self-Managed Cheaper?
For the past two years, 451 Research has published a heatmap that shows the break-even points between public and private cloud. Figure 1, the latest iteration, shows which option is cheapest, depending on utilization and labor efficiency, and is based on the CPI public cloud benchmark and the cheapest private cloud quotation we received.
Utilization is key, because private cloud is generally purchased as bulk capacity. If that capacity isn’t all used at any point in time, it is wasted, and unit costs increase. On-demand public cloud doesn’t require such capacity planning and thus isn’t as susceptible to utilization as a factor.
Green shows a net saving over public cloud, whereas red shows where public cloud has a net cost saving; the white area effectively represents the break-even point. This heatmap reflects common sense: The more the private cloud is used (utilization) and the better-managed it is (VMs per engineer), the cheaper the unit costs of each private cloud resource will be, and the more it can save when compared to public cloud.
Figure 1: Public Cloud On-Demand vs. Self-Managed Private Cloud
This heatmap no longer represents the whole truth, however, and the market has changed. Increasingly, public cloud is following a committed model: AWS, Google, Microsoft, Oracle and others reward users with discounts for commitment and loyalty. In some cases, these discounts are for compute resources only (e.g., reserved instances); in other cases they apply to the total amount the customer spends.
Some discounts are awarded based on volume (e.g., the more storage a customer uses, the lower the unit cost) or licenses (e.g., bring-your-own-license discounts on Oracle or Windows cloud database servers). Under AWS’s and Azure’s reserved instance (RI) models, for example, a consumer can save up to 50% by purchasing resources up front as capital expenditure.
What this means is that cloud now actually is sensitive to utilization: If 10 RIs are purchased for $100, but only five are used, then each RI has actually cost $20.
If an enterprise is willing to consider committing to a capacity of private cloud, then it’s equally likely to use up-front commitments to save on public cloud. As such, a fairer comparison for public vs. private cloud is the cost of private cloud against our CPI public cloud best-case pricing, which includes negotiation and commitment discounts.
Because both are sensitive to utilization, we no longer need to assess that as a contributor to determining which cloud model is cheaper. This allows us to reframe the question. Rather than asking ‘When is private cloud cheaper than public cloud?’ we can ask ‘What should I pay for a private cloud infrastructure for it to be cheaper than public cloud on overall TCO?’ – infrastructure meaning the cost of the datacenter, hardware and software.
Figure 2 shows green where private cloud provides net TCO savings over public cloud, and red for vice-versa. It uses data extracted from job website Indeed.com to derive an average cloud engineer salary of $116,000 per annum and an industry standard multiplier to represent the costs of additional benefits and workplace expenditures (x1.7). We use colocation costs extracted from our Private Cloud Index: Managed Services report. Dotted lines show quotations received for integrated solutions.
Figure 2: Public Cloud Best Case vs. Self-Managed Private Cloud for Varying Labor Efficiencies and Infrastructure Stack Cost
If initial assessments on colocation, hardware and software suggest an infrastructure price of $20 per VM, and a company believes it can operate the cloud at an average labor efficiency of 500:1, there is a good chance it will beat public cloud on TCO, as shown by the intersection of these two points being green. On the other hand, even with a fantastic labor efficiency of 1,000:1, public cloud will likely be cheaper if self-managed costs more than $30.
Open source platforms such as OpenStack are typically cheaper on the surface than proprietary or commercial software. This is irrelevant, however, if they are more difficult to operate, thus reducing labor efficiency. If an open source platform comes in at $14 per VM month, and a proprietary one at $22, open source is cheaper on the face of it.
However, if the open source technology needs one engineer for every 250 machines, whereas the proprietary technology can operate with one for every 500, then the proprietary platform should lead to a lower TCO. This reflects the importance of tooling in cloud economics, such as out-of-the-box integrations, easy-to-use interfaces, robust development tools and support for templates, which can add considerably to labor efficiency.
At fewer than 300 VMs per engineer, it is very difficult to beat public cloud on TCO. The same is true at a private cloud price of more than $28, even with a very high labor efficiency. Only one of our four integrated quotations could deliver at this price point.
Conclusions & Recommendations
When weighing concerns, such as reducing costs, outsource some management to reduce labor expenditures or protect revenue, doing it all will come at a premium. Even though the TCO benefit overall might be good news, finding budget (and justification) for such a commitment might not be easy.
Enterprises should assess where they are strong: What are the assets in terms of labor and technology? Where are the weaknesses? Then be totally up front with the service providers you talk to – don’t pay more for things you don’t need but be prepared to pay extra to offload a management burden.
It is the service provider’s job to appeal to your specific situation, so use the CPI benchmarks as guidelines for price and ask providers to justify why you should pay more.
Running your own private cloud will likely be more expensive than public cloud. For most CIOs, this isn’t a big problem if the premium is reasonable and adds value. However, it might be worth shopping around to see if savings are possible.
Labor efficiency is the key metric to be aware of: How well can you manage the private cloud and all of its aspects? If there are areas where you believe you have a weakness or where labor efficiency will be poor and expensive (e.g., security), consider outsourcing just these elements to third parties.
Partially managed clouds provide a good compromise. Most service providers today are open-minded with regards to hardware and model – tell them what you want, and if they don’t oblige, it is your business they are missing out on.
If your organization has a sophisticated in-house cloud management team and is willing to invest in developing and differentiating its services at the infrastructure level, consider the build-operate-transfer model, which uses outside professionals to architect and secure a private cloud but offers the option of transferring operations inhouse once the kinks have been ironed out.