If an on-premises server fails, IT departments cut over to a backup system immediately. But what do you do when your cloud computing service goes on the blink?

As more companies plug into the cloud, IT professionals are trying to answer that question. And fast. Last week, Google's Gmail went down for two hours, the second time in two weeks. Citrix's GoToMeeting and GoToWebinar were temporarily unavailable. A month ago, Amazon.com's Simple Storage Service was out of commission for an excruciating eight hours.

"Outages are just a reality," says Tim O'Brien, director of platform strategy for Microsoft. "Microsoft is a reputable company, Google's a reputable company, and Amazon's a reputable company. Even if you do your due diligence, you still have to manage around these risks, even if the risks are particularly low."

All of which is forcing customers to rethink their dependency on software as a service and other cloud services, and devise strategies for the inevitable breakdowns, as market researcher Harris Interactive has done. "We have some users who absolutely live and die by that database and the data in Salesforce.com," says Dan Chiazza, Harris Interactive director of global sales operations. "For them, they would have no business coming into work if it wasn't up."

There are steps companies can take to minimize the risk, including storing data with multiple service providers and regularly backing up SaaS data on on-premises servers. Forrester Research analyst Liz Herbert recommends that customers ask whether a prospective service provider has geographically dispersed redundancy built into its architecture and how long it would take to get service running on backup. Salesforce provides visibility into its system reliability at trust.salesforce.com.

Be wary of service providers with a short track record. The cloud computing market is full of startups, and while new service providers shouldn't be shunned, extra diligence is in order. Ask about their data center facilities, customer references, and any assurances they can provide that your data and applications will stay online. Archiving-as-a-service startup Casdex has gone so far as to set aside funds to ensure that its service keeps running for six months even if Casdex itself runs into financial trouble, says CTO David Barley. The Linkup, a storage service for consumers, closed down Aug. 8, leaving some customers with no way to access their data.

When negotiating for cloud computing services, IT departments should insist on service-level agreements that have some teeth to them, Herbert says. SLAs that guarantee a high level of uptime aren't necessarily standard. Salesforce doesn't commit to uptime thresholds -- unless you insist on it during negotiations, she says.

Most SLAs reimburse customers for lost service, but not for revenue lost as a consequence of service outages. Amazon applies a 10% credit if S3 availability dips below 99.9% in a month, and last month's outage forced Amazon to make good on that policy. "That's all well and good, but doesn't make the business whole relative to the damage," says Gartner analyst Rob DeSisto.