Companies such as Hewlett-Packard tout cloud service-level agreements to attract customers, but cloud SLAs do little to help companies recoup outage-related revenue
In fact, many cloud services vendors promise between 99.95% and 99.9% of uptime on a monthly basis, but typically only pay out on their service-level agreements (SLAs) if service falls far below those levels.
When Hewlett-Packard (HP) made its Cloud Object Storage and Cloud Content Delivery Network (CDN) offerings generally available last week, the company bragged a premier SLA with a guarantee of 99.95% (three and a half nines) uptime or better per month. Otherwise, it credits customers as much as 30% off their monthly bill -- but there are caveats.
In order to qualify for the full 30% credit off the storage bill, the users' service level must fall to 99.0% (two nines) or worse. If service falls between 99.95% to 99.9%, customers are entitled to only a 5% credit to their storage bill for that month.
To boot, monthly availability percentages aren't exactly straightforward. They are calculated by averaging the "interval availability" percentage of each six-minute interval over the entire month. The interval availability is determined for each HP Cloud Object Storage account during six-minute intervals as 100% minus the total number of requests that result in internal server errors or service unavailable errors, divided by the total number of requests during that six-minute interval, according to HP's cloud computing SLA website.
It's no wonder some experts said cloud service-level agreements are more of a marketing slide bullet point than a real benefit to customers.
Meanwhile, HP's storage and networking SLA credits are actually on par or slightly better than SLAs offered by similar cloud services.
For example, Amazon Web Service's SLA for its Simple Storage Service (S3) promises 99.9% (three nines) uptime and credits users with up to 25% off their monthly bills only if performance falls to 99.0% or lower.
Microsoft offers a 99.95% monthly uptime guarantee for Windows Azure Compute Services and 99.9% for storage and other Windows Azure services, and discounts users' bills by up to 25% if uptime falls to 99.5% or below.
"Obviously, HP's 30% is no big deal," said Roger Jennings, a cloud developer based in Oakland, Calif. "One-upmanship in SLA payments is a desperate move."
Perhaps more importantly, when a cloud outage occurs, credits do little to reimburse customers for lost income.
"Credits against monthly payments very seldom, if ever, recoup losses in revenue and reputation incurred from an outage of any significant length," Jennings said.
Efficient service -- not a SLA -- is what companies look for in a vendor.
"Businesses running applications in the cloud don't need or want a SLA; they want to deliver reliable service, and a SLA does not help," said Shlomo Swidler, CEO of Orchestratus Inc., a cloud computing consultancy in West Hempstead, N.Y.
Still, a cloud computing SLA does serve a role in providing an enforceable agreement that assures a certain amount of trust because, despite constant testing and maintenance, outages occur.
"Mishaps happen, networks go down and hardware fails. ... If you're going to trust a third party, you need some assurance it will be up when you need it ... [and the guarantees are] the minimum they can do," said Vernon Turner, senior research vice president at IDC in Framingham, Mass.
The good news is many cloud purveyors routinely provide at least availability above 99.95% when measured over a three-month timeframe, according to the uptime monitoring site CloudHarmony. In fact, a majority of cloud sites tracked demonstrated five nines (99.999%) -- and even 100% uptime -- on a 90-day measurement basis.
Among services with 99.999% uptime as identified by CloudHarmony in early August were Amazon's S3 service for the U.S.-East region. Akamai CDN had a perfect 100.000%, as did HP's Cloud offering. (Ostensibly, the new object storage and CDN are too new to be rated.)