Service-level agreements are the bedrock of cloud business. Providers draft them to reflect their business models,...
and customers sign them -- often without too much review -- hoping they will get what they paid for.
Businesses look for cloud providers to manage the reliability and availability of their applications and data. Depending on whether the offering is platform as a service, infrastructure as a service or software as a service, there will be differing levels of responsibility for the cloud provider. Ultimately, though, the customer is primarily interested in ensuring availability and security.
Negotiating an SLA in cloud computing, however, isn't always easy or possible, said Dave Goodman, a director at Alsbridge Inc., a Dallas-based management consulting firm. What a public cloud provider will likely offer is:
- Monthly compute availability SLAs at 99%, perhaps even pushing to 99.95%.
- Availability percentages that are typically non-negotiable, based on estimates created from the provider's underlying infrastructure reliability. Maintenance windows are normally excluded from the SLA.
- A multifault SLA that involves at least two fault domains, zones or sets. For the provider to violate the SLA conditions, both fault domains would have to fail.
- SLAs that cover network availability and performance, service responsiveness and other service aspects. Very few providers have SLAs for compute and storage performance.
"As far as a negotiable area, the primary space I've seen this in is in pricing, and it's normally associated with significant upfront commitment from the customer -- with a penalty if they don't meet the commitment," Goodman said.
Most cloud providers are only willing to issue credit if they don't live up to their SLAs, said Dan Conde, analyst at Enterprise Strategy Group Inc., an analyst firm in Milford, Mass. "I don't think they will ever give you money back, but if you have some flexibility and can deal with downtime, the credit is nice," he said.
The negotiating power an organization has over an SLA in cloud computing often depends on its size.
"I think smaller customers, especially at the self-service level of cloud purchase, are beholden to the standard terms offered by the provider," Conde said. However, if you are a giant, such as Netflix, or an organization that can purchase a larger block of services, you might be given more flexibility. Users that have a dedicated account representative from the cloud provider might also have some wiggle room.
The difference between cloud and traditional hosting services, which offered more models and choice, is cloud users consume something that is standardized, said Liz Herbert, vice president and principal analyst at Forrester Research. "Public cloud vendors won't create a separate architecture, because that dilutes the value of their model," she said.
But the market may be taking tentative steps toward greater flexibility in the form of "T-shirt" programs.
"IBM has told us their public cloud has three sizes you can choose from. As you go up in size, you get more uptime. In other words, based on what you pay, you can get more," Herbert said. Even within those differently sized environments, IBM allows you to pick and choose different characteristics around disaster recovery, performance and uptime.
Other vendors, such as Oracle, are choosing a similar path as a way to apportion the costs of higher uptime to those willing to pay.
Cloud storage SLAs specify uptime guarantees, but not data availability
Cloud storage SLAs provide uptime guarantees, but don't offer many specifics on data availability and data protection. Users can try to customize the SLA for cloud backup and archive. Terri McClure, a senior analyst at Enterprise Strategy Group, talks about what end users can expect to find in their cloud storage SLAs.
Even if you can't negotiate an SLA in cloud computing, read the fine print and make prudent choices to yield benefits. Organizations can get hung up on performance numbers without really having a baseline for comparison, Herbert said. Furthermore, many issues with cloud availability are really the fault of public internet latency.
"Often times, it isn't the cloud provider that's the problem -- it is internet bandwidth," she said.
Finally, it's important to be realistic. An uptime SLA in cloud computing might not be your ideal, "but they are often better than what customers are able to achieve in their own data centers," Herbert said.
Cloud SLAs need to change with the times
Create an effective cloud SLA
Break down a cloud SLA with this essential guide