peshkova - Fotolia

Don't get hoodwinked by cloud SLA promises

Although service-level agreements are important, desktop as a service vendors often include fine print that minimizes their fiscal penalty. IT shops should focus more on the provider's reputation.

Companies interested in desktop as a service comb through service-level agreements before signing on the dotted line to ensure apps and desktops are available when they're needed, but all those crossed T's and dotted I's may not actually matter.

Many desktop as a service (DaaS) providers promise 99.9% uptime, but limitations written into service-level agreements (SLAs) save vendors from paying out sizable service credits, if any at all.

For example, a cloud provider might offer a 99.995% SLA, but it builds in 30 minutes to fix an issue. And service-level agreement credits aren't applied automatically; customers have to take a number of steps, sometimes within a short time span, to actually reap any credits.

More often than not, the monetary compensation built into SLAs is so limited that it doesn't offset the losses caused by performance issues or outages, said Jason Read, founder of CloudHarmony Inc., a cloud service analysis firm recently acquired by Gartner that tracks cloud outages.

"It seems like the SLA would help, but if you experience an outage, it really doesn't," Read said.

For example, VMware's Horizon Air Desktops SLA states VMware will use commercially reasonable efforts to ensure each Horizon Air Enterprise Center Console instance -- and all virtual desktops managed under that instance -- is available during a given calendar month equal to the 99.9% availability commitment. That amounts to 10.1 minutes of downtime per week or 8.76 hours per year.

Like many cloud provider SLAs, VMware's uptime commitment excludes "force majeure" events, denial-of-service attacks, viruses, hacking attacks or any other events that are not within VMware's control, along with other exclusions such as scheduled maintenance where customers are notified 24 hours in advance. If the availability of the instance is less than the availability commitment, IT can request service credits for that instance.

The SLA for dinCloud Inc.'s Hosted Virtual Desktops guarantees the service is available 99.95% of the time -- which allows 4.38 hours of downtime per year or 5.05 minutes per week -- in a given billing cycle, barring planned maintenance. If the company fails to meet this guarantee, customers receive a credit.

DinCloud alerts customers of disruptive system maintenance changes at least three days in advance, but emergency maintenance is exempt from the SLA's service credit promise. Service credit is the percentage of the monthly service fees for the service that is credited to clients for a validated claim.

Each cloud provider also has its own timetable for incident reporting. To be eligible to submit a claim to dinCloud, the client must first notify dinCloud's client support of the incident -- using specific procedures -- within five business days following the incident.

Other providers allow a bit more time to submit claims. Amazon Web Services (AWS) users who experience an Elastic Compute Cloud (EC2) outage have until the end of the second billing cycle after the incident occurred. The claim, which customers enter into the AWS Support Center, must include the words "SLA Credit Request" in the subject line. It must also include the dates and times of each unavailability incident, the affected EC2 instance IDs and request logs that document the errors and prove the outage, or else it won't be eligible.

If AWS affirms a credit is due, it will be issued within one billing cycle following the month a service credit request is confirmed.

Amazon provides a DaaS service, Amazon WorkSpaces, but hasn't published an SLA for that service. The company did not respond to a request for SLA information.

Cloud SLA terms favor providers

Cloud computing SLAs may never be enough compensation in any customer's mind, but there are sound business reasons why all cloud providers limit SLAs, said Mike Chase, CTO and executive vice president of dinCloud.

"We have a fiduciary responsibility to stay in business for the benefit of all customers," Chase said. "Thus, you can't allow one incident or one lawsuit to take the ship down."

Plus, technology is bound to fail, so providers must account for that.

"Part of the compensation in an outage is the fact that a customer can tap their feet waiting for the cloud to come back up while we are the ones stuck with rousting engineers out of bed, stocking millions of dollars' worth of spare parts, paying a small fortune in vendor support contracts, top-tier data centers worldwide like Equinix and much more," Chase said.

"The question becomes, do we hand a big chunk of money to the customer or do we fix the problem, and when things are OK in the cloud [can] they get that super-low-cost structure they couldn't get without cloud?" he added.

There isn't a correlation between cloud SLAs and performance.
Jason Readfounder, CloudHarmony Inc.

Indeed, cloud providers advertise SLAs they are willing to meet, and they have teams crafting those SLAs to limit exposure, said David Bartoletti, an analyst with Forrester Research. The secret to cloud success is to match workload requirements and expectations to what the cloud provider offers, Bartoletti said.

"You don't go to public cloud to demand a customized SLA. You match the right workload to what the cloud advertises," Bartoletti said.

IT shops that require specialized SLA terms, such as higher total availability than 99.995%, shared risk beyond refunds or credits, or shorter recovery point objective for data failure may find a managed services provider (MSP) or hosted private cloud provider a better fit for their desktop hosting needs. MSPs and private cloud providers are willing to craft tighter SLAs and share greater risk than the commodity public clouds, Bartoletti said.

"You can always craft higher SLAs with someone willing to charge you more for it," he said.

In general, if DaaS SLA compensation were high, benefits -- particularly in cost savings -- would be low, Chase said. Nonetheless, even a small service credit payment set across enough customers can be a huge hit to cloud providers.

"It is always a wake-up call for anyone in the cloud to say 'how can we do better?'" Chase said.

Types of SLA policies

Though cloud service credits typically don't amount to much, it's important to understand how a DaaS provider calculates downtime and service credits.

Some cloud providers pay credits based on a prorated length of downtime that exceeds the service agreement. Credit is issued based on that calculated amount, capped at 100% of service fees.

For instance, with VMware Horizon Air desktops, if availability falls short of 99.9% in a calendar month, customers can request one SLA credit, plus one additional SLA credit for each additional 300 minutes the instance was down, up to three service credits per month. An SLA credit is capped at 10% of the monthly service subscription. If an instance is down for 24 consecutive hours, VMware customers can request a Chronic SLA Credit, which equals 100% of the monthly service subscription and metered charges in a month.

Another type of SLA is threshold-based, which means credits are not redeemable until an outage exceeds a given time. There are also percentage-based credit policies which discount future invoices based on the amount of downtime and the terms of the SLA.

The most fair and simple policy, according to Read, appears to be the prorated method. The threshold method gives cloud providers the greatest protection and flexibility because most outages tend to be shorter than the thresholds, according to CloudHarmony's data.

Vendor Math

Availability in a given month is calculated according to this formula: Availability = ((total minutes in a calendar month - total minutes unavailable) / total minutes in a calendar month) x 100.

SLAs no barometer for performance

A 100% SLA does not mean the provider will actually deliver non-stop availability. Service commitments don't actually correlate to uptime, according to Read.

Many cloud vendors with the worst uptime offer 100% SLAs, whereas some of the top cloud providers have the lowest SLAs, according to Read. For example, Amazon EC2 promises 99.5% and Linode LLC, which has dismal uptime performance, promises 99.9%.

"The biggest takeaway is that there isn't a correlation between cloud SLAs and performance," Read said in a recent interview.

Other factors, such as historic availability, outages and the reasons for those outages, management processes and the outage post mortems, can give IT a better feel for how well the provider performs, he said.

Next Steps

A deep dive into evaluating cloud SLAs

Five things to consider in a cloud SLA

Breaking down SLA availability versus performance

Dig Deeper on Cloud-hosted virtual desktop and application strategy