Getting the cloud you pay for

Read time 11min 40sec

Personal cellphone contracts can be difficult to choose, among the many on offer. If a hapless user selects a contract ill-suited to her needs, the bill can spiral beyond her budget. Switching from a pay-as-you-go option to a fixed-term contract can greatly influence monthly costs.

The bills for cloud computing also depend on the way the service provider charges the customer. Subscription-type billing has been the norm in SA. Recently, billing engine technology to charge only for resources actually used became widely available. The majority of local service providers appear in no hurry to implement this riskier approach, however, citing an immature market. Many customers are ill-prepared to understand even the basics of their real IT costs, after all.

A local tier-one private cloud service provider has seen some demand and offers usage-only utility billing. But it manages the significant risk by understanding customer usage patterns thoroughly beforehand and giving them incentives to smooth out demand themselves.

Paying for peaks

True cloud services should bill the customer only for resources actually used, according to standards body NIST. But South African providers don't do that. Instead, most are charging customers according to subscription-type models on the infrastructure-as-a-service (IAAS) layer.

"The challenge in SA is that we don't have pure-play cloud providers yet, because the market hasn't quite reached that stage," says Gareth James, solution strategist at CA Southern Africa, a technology provider.

"We have hosters and telcos fighting over that space."

Adds Rainer Jeske, MD of technology solutions at distributor Datacentrix: "All the Microsoft Exchange scenarios that are sold at the moment are sold on a cloud service catalogue basis, where the customer is billed per month, per mailbox."

Current subscription billing models can be divided into two main groups. In the first, the customer pays the provider to reserve a certain number of resources - reserved instances - and gets billed for all those resources, whether used or not, at the end of the month.

In the second, a customer starts off with a zero account at the start of the month, provisions cloud resources as demanded and gets charged at the end of the month.

"Today, most cloud providers give you an on-demand price and a reserved price with some incentive for additional capacity you may need," says JJ Milner, MD at Global Micro, a service provider.

"As an example, say you buy a cellphone package with 500 included monthly minutes. The 500 minutes are like your reserved cloud resources. Any extra minutes are billed at the out-of-contract rate, like your cloud computing on-demand price."

We don't have pure-play South African cloud providers yet.

Gareth James, CA Southern Africa

However, providers do not charge for resources used only, but for the 'high water mark' - the peak usage - during the month. To understand which cost centres used the most cloud resources, customers will usually have to develop custom reports themselves.

But telco and service provider Vodacom claims its self-service, scalable IAAS public cloud service can provide aggregated reporting at any level in a customer's cost centre hierarchy. And that this reporting can also be extended to the customer's private cloud.

"We built the platform ourselves, but adopted the Novell cloud manager. Business people provision resources on the fly," says Richard Vester, head of hosted services at Vodacom Business. "You can break down the provisioning reporting by company, division, branch, or user, and then roll it up at any level in a monthly or annual report.

Just usage

A third approach, where a customer is only billed for actual resources used, has been possible for some years at one vendor, but only become widely feasible this year.

For the past five years, utility-style metering at the IAAS, PAAS and SAAS levels (platform-as-a-service and software-as-a-service) has been 'do-able' on an Oracle implementation, says Andrew Bond, technology director EMEA at Oracle, a technology and solution vendor. It requires work beyond solution installation, though.

At competitor IBM, the mature Tivoli offering provides granular metering, billing and reporting to the business's cost centre hierarchy, as part of its IAAS and SAAS-level solutions.

"Other providers struggle," says Werner Lindemann, VP of systems technology group at IBM, "because they don't have the sophisticated software that can do the systems management as well as billing and metering".

Recently, cloud utility-billing became far more widely available.

"It is only since April or May 2011 that some of the more open systems from providers such as HP, Microsoft or CA allow you that fine level of usage metering," says Jeske, "as well as feeding it back into a billing engine".

Edgier clouds

On the face of it, it would always be better to only pay for resources used, rather than peak usage during the month. But this approach is also riskier for both service provider and customer.

For the service provider, usage-only billing removes the financial safety net provided by subscription-type catalogue services.

Continues Jeske: "With cloud service catalogues, a company can calculate what capacity data centre to build, and what to charge its customers to recoup the cost. Such services are easily established, and there is no risk on both sides. The customer knows exactly what he has to pay and the service provider knows what he is in for.

"But with usage-only billing, the cloud builder takes the risk: he promises certain capacities to customers, but the customers only charge for actual usage. So if the cloud builder constructs a data centre, but end-users only use half the capacity, the builder has invested far more than he should have."

Confirms Milner: "There is nothing for nothing in this world. In some cases, being a cloud provider comes with a lot more responsibility and business risk. Sometimes, this model drives in new costs that didn't exist when you are only handling 10 000 users. When you handle a million users, you need three data centres and five telcos behind you."

The risk can be big enough to put a service provider out of business.

"Service providers can't afford to do pure, usage-based metering because it is too unpredictable," says Greg Hatfield, GM of cloud services at Dimension Data.

"You charge for the average, and then you get a handful of people that exceed that. Margins are thin in cloud because of the scale expectations. With utility billing, you could be unprofitable overnight."

Higher unit costs

Apart from having one's cloud provider possibly going out of business, customers also face the risk of possible higher per-unit costs with a usage-only billing model.

"I have quite a few discussions with customers about this. There could possibly always be a bit of an overcharge to the customer, simply because the provider has to cater for the costs somehow," notes Jeske.

"Only massive data centres like Google can have 10% or 20% spare capacity at any one time, and sell that for whatever they can get."

"Consumption-based services are often synonymous with cheap, and the way you get cheap is you get scale, and the way you achieve scale is you deny people the ability to customise," adds Hatfield. "You also potentially cut corners and try to lower your unit cost. All of these factors are diametrically opposed to what the enterprise is looking for in a solution. At the enterprise end of the market, service level agreements, data security and privacy, your operating model and the sustainability of your business - various strategic factors similar to those you would consider for an IT outsource - are more important to clients than saving a buck here or there."

Cloud metering and billing is complex, no matter what approach is used. Management systems in the data centre have to feed back metering usage information to the billing engine, which then constructs the customer's bill.

To enable usage-only metering, the service provider has to add a new system management layer to the cloud computing architecture. Further complicating matters, industry standards for metering and billing are still emerging, as well as standards that would make it simpler to move from one cloud provider to another.

Cloud utilities

Some cloud providers do not see demand for usage-only utility billing in SA. But in March, tier-one solution and service provider T-Systems introduced its hosted private cloud PAAS service, with usage-only billing at the IAAS level, in SA. It is a rapidly elastic, measured, on-demand self-service.

"Our Dynamic Services solution is pure usage-based," states Roelof Louw, pre-sales at T-Systems SA. "We monitor the processing units and storage used, and invoice that at the end of the month. Usage is measured on an hourly or daily basis, depending on the exact setup of the system. You pay for the actual amount you used up to the time of the measurement. If you use 5GB of storage per day, and at month-end you use 50GB, then for 29 days you pay for 5GB, and on the 30th day you pay for 50GB."

Other providers struggle because they don't have the sophisticated software that can do the systems management as well as billing and metering.

Werner Lindemann, IBM

As with other cloud providers, the customer needs to fit into one of the standardised environments, or images, T-Systems offers. That means choosing one of the combinations of applications at specific versions and patch levels.

Continues Louw: "If you come to me and say you want SAP version 4.6 on service pack 2a, a very specific customised solution, then I can't do it. One of the big characteristics of cloud computing is this standardisation. But our solution is very flexible - up to the infrastructure provisioning point - for easily deploying a new system and scaling resource capacity."

As part of the solution, customers can draw reports on the billing according to their cost centre hierarchies, much like they would for utilities like water and electricity.

"The monitoring, measuring and tracking solution in Dynamic Services gives us information on what resources were used in the infrastructure layer," explains Louw. "In the implementation process, we translate that information to the customer's cost and cost centre structures."

The service offers rapid elasticity with its self-service resource provisioning, but it is limited by the provider to some extent. There is no such thing as a sudden 300% increase in demand from a customer, says Louw. T-systems manages customers in two major ways.

Getting to know you

First, it gets to know the customer thoroughly.

"When we sign up a customer, we get to know their environment and applications, and look at the average historic usage. Our commercial engagement model is based on this," says Louw. "Either the customer already has historic information available and we can base the contract on that, or we do a cloud-readiness assessment, where we measure peak and average usage over a period.

"We have a very stringent application migration process. We do a lot of assessments before moving a customer's application from their current infrastructure to ours on Dynamic Services. It means understanding peak and trough demand, the trends and the growth patterns.

"The customers have normal usage peaks: at month-end, you'll have 20% or 30% more utilisation. We've been doing this since 2004. We don't have customers that have enormous usage peaks of 50% or 100% or more that we can't foresee or predict."

Margins are thin in cloud because of the scale expectations. With utility billing, you could be unprofitable overnight.

Greg Hatfield, Dimension Data

Secondly, pricing agreements give the customers incentive to smooth out their demand themselves.

"It is a commercial model. When we sign the customer, we limit them to a certain percentage in increase and decrease. Above that, we bill at a premium, for the peak usage," continues Louw.

"We also put a bit of a premium on usage if a customer only uses the IAAS service for a weekend, or for two or three days, which makes the environment difficult to control and manage. Adding the premium makes it much easier for us; we don't get a lot of peaks of very short duration.

"We must have capacity available for peak customer demand, but when it is not used, I can't sell that capacity to someone else. The more the customer can predict peak and trough demand, or we can control that, the less of a premium I have to charge."

The future is public

Utility-metering and billing has recently become widely available at the IAAS level, but few service cloud providers are offering it so far. As in any 'thin margin' infrastructure business, there are significant risks for service providers in covering their costs. A provider that does offer tier-one usage-only billing aggressively manages its commercial risk, first assessing customers stringently, then giving them incentives to smooth out demand themselves.

The vast majority of cloud computing is still private cloud, an estimated 90%. But a shift within five years, so that 60% will be in public clouds, is expected.

At the SME end of the market, a demand for usage-only billing is already emerging, primarily driven by a need to understand cloud costs before moving there.

Most South African cloud providers will likely stick with subscription-style billing in the medium term. That way, both provider and customer can sleep easier, knowing that the cash flow for the cloud bill is likely to be available.

See also