451 Research is recognized as a global leader in a number of areas – including, and especially, in tracking the cost of cloud via its Cloud Pricing Index. In this report, CPI architect Owen Rogers and CPI analyst Jean Atelsek apply their methodology to private cloud total cost of ownership (TCO).
TCO is a complex issue, and Rogers and Atelsek do an excellent job of defining ratios that businesses can apply to understand their private cloud position, the options (ranging from DIY to hosted) that may make sense for them and the keys to obtaining best return on private cloud expenditures. The benchmarks themselves may shift across contexts (and over time), but the 451 approach to defining private cloud options and issues will be beneficial to all firms looking to optimize private cloud spending (ed).
This year’s private cloud index (pCPI) finds a market for cloud, where public, private, managed and unmanaged options can all be the best option when it comes to total cost of ownership (TCO). How can buyers understand the options with such complexity? Our guidelines can help.
The 451 Take
If your organization is focused on efficiency, then private cloud might be the lower TCO option. Centralized applications, just-in-time hardware provisioning and prioritized workloads can drive high utilization, while automation, homogeneous workloads and outsourcing can drive labor efficiencies. Here, OpenStack is good value. But for those without the means to operate in this ‘Goldilocks zone,’ stick to public cloud if price is your major concern. But if you’re willing to pay a bit more, on-premises managed offerings or the stalwarts VMware or Microsoft are probably your best bet. But don’t rule out hosted managed options either – on average, they cost 15% more than public cloud, but it all depends on the provider. Hosted options are often bundled with other services, including security audits and enhanced monitoring – but value-added is only valuable if you need it. There is no simple answer and buyers will need to do their homework. But this homework shouldn’t just consider today; it should examine how the business will change, and what impact this will have on the IT estate, specifically in terms of utilization and labor efficiency. And furthermore, what if the business doesn’t evolve as expected?
Introduction
Private cloud has long been touted as the premium choice – an ‘enterprise grade’ cloud for mission-critical requirements but with a premium price. Data protection, ownership and control are the typical arguments for going private pushed in the media and by those advocating its underlying technologies. When it comes to TCO, many consider public cloud to be the standard choice for the cost-conscious CIO – no surprise, considering the frequent price cuts announced by the hyperscalers and the resultant press attention they draw.
However, private cloud can be a less expensive option for enterprises than public cloud. Forty-one percent of 150 IT decision-makers surveyed in February 2017 as part of a custom 451 Research project for VMware claimed to be operating their own private clouds at lower unit costs than public cloud. An additional 24% of those surveyed said that they are paying less than a 10% premium for their private cloud.
But cost reduction against public cloud is not the primary driving force behind private cloud adoption. In the same survey, protection of data was the most popular benefit of a private cloud, with asset ownership and integration with business processes also ranked highly. However, cost benefits do still matter: more than half of respondents rated cost efficiency as a key driver of the decision to use a private cloud. So public cloud is usually cheaper, but certainly not always.
As such, service providers and vendors that claim they don’t win on price are showing a lack of respect to their potential buyers. It is true that the cheapest solution isn’t necessarily the one that end users buy, but it’s arrogant to assume that end users aren’t doing their homework and weighing the costs and benefits of different options against each other. Never has the market for infrastructure been more competitive.
But while buyers are happy to have conversations with providers and vendors, and certainly want a ‘trusted adviser’ to help them maximize their use of private cloud, they do have a challenge – when should they use private cloud, and when should they use public cloud? After all, buyers want to access those ‘best execution venues’ (BEVs) that meet their performance, cost, contractual and other requirements. Comparing these options is difficult – not just in terms of technical capability, but also in terms of bare-bones financials. These options are priced, publicized and invoiced in entirely different ways.
The Cloud Price Index: Private Edition (pCPI) is designed to help draw back the veil by deriving the average price of a private cloud using a ‘basket of goods’ approach. By considering the total cost of a bundle of hosting services, infrastructure, software and operating systems from a range of providers delivered using a range of mechanisms, we can find average prices per VM-hour and GB/month for our compute and storage requirements. Buyers and sellers alike can use this information to better determine pricing and procurement strategies.
Of course, it’s not this simple in real life – there are factors such as migration, existing investments, compression, resiliency and millions of reasons beyond price and TCO why an end user might choose one option over another. We don’t seek to provide answers to all these questions or issues. The aim of the Cloud Price Index (CPI) is to provide a basis from which end users and service providers can make assessments of what is reasonable, both in terms of price and ultimately value.
Key financial decision criteria
The unit cost of a virtual machine (or any reason) running on a private cloud comes down to two factors: labor efficiency and utilization. The greater the number of VMs an administrator can successfully manage (i.e., its labor efficiency), the lower the unit cost per resource. The better-utilized the private cloud (i.e., its utilization), the lower the unit cost per resource. As such, the financial gain or loss in using public cloud over private comes down to these two decision criteria.
To make a TCO-led decision, enterprises need to understand and predict the utilization and labor efficiency of their own private clouds to understand if it can compete with public cloud. Different levels of utilization and labor have different optimum configurations. Of course, there are other drivers and concerns beyond TCO, but here we focus purely on price.
Labor is a truly critical factor in TCO, but perhaps one of the most difficult to measure. Imagine a perfectly automated, completely self-service private cloud. In this dream-like scenario, hardware could be added and swapped out automatically and robotically, patches could be tested and applied with no human involvement, policies could be configured and enacted routinely, and network, groupings and server pools would ‘just happen.’ Self-service would remove many of the tasks required of the administrator. In this case, an engineer could manage a vast array of hardware. After all, the orchestration layer is doing most of the work in this scenario, and the engineer would need to become involved only when the unexpected takes place. So the question is how beneficial is a new technology or service in reducing the management overhead?
From a custom study conducted by 451 Research for Hewlett Packard Enterprise’s Flexible Capacity, the average labor efficiency of a private cloud was 464 virtual machines per engineer, but it will be different for each company and application. Potential buyers should aim to quantify labor efficiency today and in the future. What difference will outsourcing, tools, automation and self-service make to labor efficiency? Understand the tasks your administrators are performing today, and be realistic of how efficient they will be able to be in the future. In public cloud, theoretically, labor requirements should be close to zero when it comes to infrastructure management because it was outsourced to the cloud provider.
Utilization is a measure of what percentage of a private cloud is being used for activity that provides value to the organization. In an ideal world, a private cloud would have 100% utilization – every single cycle, bit and electron would be used to perform a valuable task such as running an application. Unutilized capacity is effectively waste; resources that are not being used all the time are sunk costs. If utilization is high, the VM cost can be significantly lower.
From the above HPE study, the average utilization of the 150 enterprises’ decision-makers surveyed was 58%. Potential buyers should aim to quantify utilization today and in the future. What difference will variability of application demand, forecasting data, compression and allocation tools make to utilization? Understand and forecast how demand will change, and run what-if analyses to determine your risk exposure. In public cloud, utilization should be close to 100% when it comes to infrastructure management because on-demand pricing means it’s the provider’s problem, not yours.
The public cloud pricing index has seen pricing slide downward for the past three years on nearly every service covered. There is little evidence to suggest private cloud has followed the same path during the time. Enterprises should plan for public cloud to continue to decline and thus private cloud to be relatively more difficult to make cost benefits over public cloud.
A question of scale
The findings of this year’s Private Cloud Index revolve around two scales. It’s not an exact science and they are guidelines based on our specific scenario, but they provide benchmark prices. Full details are in Cloud Price Index: Private Cloud Edition Q3 2017 and The Great Public vs Private Cloud Debate 2017, including actual benchmark prices for all options based on utilization and labor.
- Standard scale, where the private cloud is anticipated to be operating at less than 400-500 VMs/engineer @ 100% utilization (or 800-100 VMs/engineer @ 50% utilization).
- Commodity scale, where VMs are cheaper than public cloud, but where labor efficiency must be greater than 400-500 VMs/engineer @ 100% utilization (or 800-100 VMs/engineer @ 50% utilization).
The figure below shows these scales, with the standard scale (where public beats private) shown in blue and the commodity ‘Goldilocks’ zone shown in red. In the Goldilocks zone, conditions are just right for private cloud to beat public cloud – just as the baby bear’s porridge was just the right temperature for hungry Goldilocks in the fairy tale. The depth of color shown at the intersection of any labor efficiency and utilization shows the relative advantage of going private over public.
Standard scale
In standard scale, chances are public cloud is going to be the cheapest option. But life is not that simple, and many enterprises are willing to pay a premium for the single tenancy, security and performance associated with private cloud.
For those that want private cloud, on-premises managed offerings have significant TCO benefits. On-premises managed is a newer cloud model where a third party handles most cloud and orchestration administration, while the user is responsible for colocation and a degree of hardware management/swap-out. On-premises managed offerings provide remote management support so that buyers can outsource management of the hypervisor and orchestration with monitoring and other features included, but the offering can be hosted in the user’s datacenter. Our analysis suggests it has significant TCO benefits.
For those that want to manage their own, commercial offerings (which we define here as Microsoft and VMware) remain the cheapest TCO option. To achieve the same price as commercial offerings, an OpenStack platform would need to allow management of a minimum 12% more VMs than a commercial offering using the same number of engineers. Considering the maturity of the tools included with commercial platforms to aid automation, scheduling and management, we think it is unlikely that OpenStack can provide a 12-18% management improvement benefit over commercial offerings when it comes to labor efficiency.
However, if OpenStack is the priority for reasons such as portability and interoperability, distributions have a clear benefit over using a DIY approach. To make a distribution worthwhile, it should provide a small labor efficiency benefit of about 10% compared with DIY. We think distributions can easily achieve such a benefit, considering the tools for installation, labor and efficiency that are bundled with them.
Hosted managed cloud, on average, is about 15% more expensive than public cloud. However, some private cloud quotations we received were cheaper than public cloud quotes, suggesting buyers can get the best of both worlds – single tenancy at public cloud prices.
In general, the more features and services that are included as standard in a managed private cloud, the more expensive it is. This suggests the market is still value-driven, despite private cloud being ubiquitous today and comparative public cloud pricing being readily available. Security is the key value-added area of services, being 4x more likely to be included in more expensive private clouds than cheaper ones. So hosted managed cloud should still be considered by buyers, but it’s important to factor in all the value-added extras. And remember: just because something is bundled in, doesn’t mean it’s good value – it’s only valuable if you want or need it.
Commodity scale: the Goldilocks zone
In the Goldilocks zone, utilization and labor efficiency are ‘just right’ for private cloud to beat public cloud. From the HPE report, 40% of organizations claimed to be operating at a price cheaper than public cloud.
In this zone, OpenStack has real TCO benefits, demonstrating its take-up among larger enterprises. Essentially, it is easier to achieve a low cost using OpenStack than it is using a commercial offering at this scale. OpenStack rapidly catches up to commercial platforms for higher levels of labor efficiency, outflanking it at a golden ratio of 600:1. As an example, to achieve a price point per VM of $40, a labor efficiency of about 800:1 would be necessary for a commercial offering. For an OpenStack distribution, an efficiency of 780:1 is needed, and 600:1 for a DIY approach.
The question is will a distribution help an end user increase the number of VMs managed by an engineer from 600 to 780 – a difference of 30%? We think yes, due to the benefits of easier installation, patching and management. As labor efficiency increases, the benefits of using OpenStack also rapidly increase relative to all other options, as a result of lower recurring licensing costs.
Private cloud does have significant risks. When a self-managed private cloud is procured, it is designed based on assumptions of utilization and labor efficiency. But the reality might be quite different due to shifting demand, use cases and applications – factors often beyond the administrator’s control. Get it wrong, and those virtual machine costs increase exponentially.
Very interesting that so much of this analysis returns to labour efficiency – a factor that is out of scope for the provider, but a key input in the CIO's calculations. In our TCBC best practices document, Cloud Economics for the IT Practitioner, the community also advises end user organizations to look first to the organization's human assets, to asses internal capabilities before making decisions on cloud deployment models.