When demand for an application suddenly spikes, IT folks make a mad dash to optimize resources, balance loads and add more hardware. And now, they're considering another option -- the cloud.
The marshalling of public cloud resources in response to usage spikes is a practice sometimes referred to as "cloud bursting." In this scenario, organizations pay only for the resources they use and forestall additional infrastructure purchases. But with most applications firmly ensconced on internal infrastructure, how does one even get there?
For some enterprises, the answer lies in their application delivery controllers (ADC), the load balancing tier already used to coordinate load across available infrastructure. A load balancer is the best tool to figure out when you need more capacity, said Tim Maliyil, founder and CEO of AlertBoot, a software company based in Las Vegas.
Maliyil speaks from experience. Last year, the 10-year-old company needed to refresh its hardware, but Maliyal didn't want to spend the money. Instead, the company moved its entire operation from its colocation provider to two geographically disparate public clouds from SunGard and Opsource, saving an estimated $85,000 a month, while continuing to grow the business.
"Our infrastructure is sitting in a closet right now," Maliyil said. "I don't know what to do with it -- maybe donate it to charity?"
Taking cues from the application delivery controller
AlertBoot used an application delivery controller from Zeus Technology, which was recently acquired by WAN optimization player Riverbed Technology. Unlike most load balancing tools that run on dedicated hardware, Zeus ADC runs as a software-only virtual appliance, which enables it to run in the cloud. The appliance also includes scripting capabilities to automatically spin new servers up or down, in response to load.
That pays off, Maliyil said. For instance, traffic to AlertBoot's point-of-sale software doubles during the holiday season. But because Zeus can quickly spin servers up or down, AlertBoot's cloud bill only increases by 30%, because the traffic management app spins down servers at night when stores are closed.
Early adopters of the bursting model are, not surprisingly, cloud service providers themselves. RackSpace built a virtualized bursting environment for its customers that used F5 Networks's BIG-IP traffic management tool to monitor bandwidth and determine when to spin up new cloud computing environments.
Next steps for application delivery
To make cloud bursting happen, ADCs must develop their automation capabilities -- or forge relationships with the major players, said Joe Skorupa, an analyst at Gartner Inc.
"The [ADC] is a very good monitoring and control point for telling us that application performance isn't what we need," he said, "but that's just the beginning."
From there, the next step is to apply that information to what Gartner dubs the "fabric resource pool manager," i.e., orchestration and automation engines such as Hewlett-Packard's Cloud Service Automation, from its Opsware acquisition, BMC's BladeLogic Automation Suite, and Dell's Advanced Infrastructure Manager.
Cloud providers themselves, meanwhile, see other impediments to cloud bursting: logistics.
BlueLock is a cloud hosting provider that caters to VMware-based enterprises, providing capacity for greenfield projects, said Aaron Branham, director of technology at the company.
Our infrastructure is sitting in a closet right now. I don't know what to do with it -- maybe donate it to charity?
Tim Maliyil, founder and CEO of AlertBoot
"Customers use us for projects that they don't know are going to work," Branham said, augmenting their internal resources with BlueLock capacity. Customers have been asking about burst capacity as of late, he said, but there are obstacles to overcome.
"I'd like to get there," he added, but "the bursting model doesn't work for us because of contracts."
Specifically, the large enterprises that BlueLock targets tend to have complex billing processes in place, and it can takes months to bring a new vendor into the fold. By then, any burst capacity the enterprise had consumed was probably spun down long ago, and it can be hard to extract payment.
"It's the pump and run scenario, where you pump the gas, but don't pay later. There's a certain fear of that," Branham said.
However, more enterprises using cloud management tools like VMware vCloud Director could lay the foundation for further cloud bursting, Branham said. As a plug-in for VMware vCenter, vCloud Director allows administrators to securely provision virtual workloads across internal private clouds as well as public cloud providers like BlueLock.
Until then, the closest BlueLock gets to cloud bursting is to provide disaster recovery services. "Instead of buying a whole second disaster recovery site, buy a 'warm site' on the cloud," with reserve capacity in case a disaster comes to pass, Branham added. That's not quite the dynamic, pay-as-you-go Nirvana espoused by users like AlertBoot's Maliyil, but it's a start.