Some might say that San Diego-based Amylin Pharmaceuticals was pushed into the cloud. The 23-year-old company's data center had been running at 100% capacity and IT executives feared inevitable performance problems, but the company was reluctant to build another site due to the poor economic climate.
"We had already done significant virtualizing of our servers and couldn't cram anything more onto our data center infrastructure. The only thing we could do to relieve the pressure was to start moving legacy applications to the cloud," says Todd Stewart, senior director of IT operations.
Some applications, such as e-mail, were switched to a software-as-a-service model so that IT could offload the application and infrastructure management to a service provider. However, there were other applications, including human resources and enterprise monitoring, that Stewart and his team wanted to closely manage without having to deal with the underlying infrastructure. To handle this delicate balance, he turned to infrastructure as a service (IaaS) through the public cloud.
With public cloud IaaS, organizations pay per use or per cluster of resources for an external cloud service provider to host their virtual servers. These types of services are available from Amazon, Google, HP, IBM, Microsoft, RackSpace, Sun Microsystems, Terremark Worldwide and telecom companies, among others.
IT maintains control over the applications without worrying about configuring, upgrading or patching servers and other infrastructure. If a department needs to deploy a new application, IT simply loads that application onto the service provider's virtual server and the software is available to users.
Infrastructure-as-a-service checklist
* Cost comparisons that include maintenance, heating and cooling
* Software licensing
* Security safeguards
* Latency/performance issues
* Contractual safeguards
* Regular reviews
In a recent report, Forrester Research found that about 25% of all enterprises surveyed plan to adopt IaaS via an external provider.
Most that go this route can expect tremendous savings, according to Lynda Stadtmueller, senior research analyst with Frost & Sullivan's Stratecast division. In a recent study, she found that a small business that migrated four of its servers to the cloud, essentially shutting down its on-premises data center, realized a 50% savings. Although it involves a small business, she says the statistic is relevant because "it's unlikely that a larger business will migrate its data center infrastructure all at once or completely."
In other words, even a larger company would migrate only a few servers at a time, so those numbers translate to what even larger companies are likely to be doing.
Savings arise from deferred costs for equipment upgrades and maintenance, power, and decreased labor costs. Most data centers provision for peak activity but operate at low utilization rates, thereby wasting valuable resources.
Bernard Golden, CEO of consultancy HyperStratus and author of Virtualization for Dummies, agrees that IaaS offers IT a lifeline when it comes to rising data center costs and real estate constraints. It offers IT the opportunity to reduce the data center management burden, yet improve customer service, he says.
Warnings about compliance, among other things
However, Golden warns that along with these positives come a lot of considerations, including how to efficiently migrate data to the cloud, handle security of data stored in the cloud, meet compliance demands, avoid software licensing snafus, test for latency problems and get applications and data back if IT executives change their minds or break with their cloud providers.
Amylin's Stewart says he and his team tackled all of these issues before moving his legacy applications into the cloud. In fact, before he even moved a single byte, he formed a "tiger team" to study the 50 to 60 services that his IT team provides and determine which were ripe for IaaS. "We looked at expenses surrounding licensing, maintenance, labor, power and infrastructure support. Then we stack-ranked each service according to its cost to see if we could deliver it via a less-expensive channel such as IaaS," Stewart says.
"Until you have a good sense of what you're spending by application, then you can't really make intelligent decisions," he says. This prioritizing also enabled Stewart and his team to show business leaders the impact of moving their applications off-site. "The business may think they want five 9s of availability, but the lower cost of an IaaS might be more appealing," he says.
Amylin's IT team identified applications, such as testing and development, human resources and enterprise monitoring, among others, that were key to the business but did not require the infrastructure to be on-site. He advises his peers to consider applications that are more back-office and less customer-facing.
Considerations beyond cost
After reviewing the cost of supporting applications in-house, the next step is to put each application through filters such as control, performance, architecture and licensing, and security. Does compliance dictate that IT must have total control over the application infrastructure? That will be difficult, since public-cloud providers rely on virtualization to keep costs low. Therefore, data will probably be moved around physical hosts as well as data centers to keep loads well balanced. If a business need dictates the requirement to pinpoint the exact location of data, then a public cloud is not really a good option.
Some of Amylin's data falls under the FDA's validation rules. Stewart says it would be too difficult to prove compliance if that data was in the public cloud and using shared resources, so he keeps that information in-house.
Chad Swartz, senior manager of IT operations at Chicago-based Preferred Hotel Group, had a similar concern when he migrated his entire data center infrastructure to Terremark's Miami facility. Swartz had chosen a wholesale move of his Citrix, SQL, Exchange, SharePoint and homegrown hotel applications because he was facing a $250,000 capital expenditure cost for a renovation of Preferred Hotel's on-site data center. His 250-employee company had already given its blessing to use Terremark for backup and disaster recovery, so he was able to make the case it would save even more money by moving infrastructure management to a cloud-based cluster of 16 physical servers.
The one catch: Because the company deals with credit card transactions, it must follow the Payment Card Industry Data Security Standard, which requires physical control over servers handling that data. To avoid any complications, Swartz makes sure all credit card transactions go directly to a third party and he avoids keeping any of that data on his outsourced infrastructure.
Performance not always a problem
Once compliance implications have been examined, the next step is to determine whether applications will take a performance hit. Software that had been living on a LAN will now be accessed over the WAN, and that will likely mean some sort of delay, Stratecast's Stadtmueller says.
"Distance always adds latency to an application. When users engage in chatty, transactional applications, they expect immediate responses. Network delays of just a few milliseconds can cause them to be frustrated," she says. In a public cloud environment, "IT has little or no control over the network path, and that makes it dicey for some latency-sensitive applications," Stadtmueller adds.
But cloud doesn't always mean slower. Jason Harper, vice president of technology at Morgans Hotel Group in New York, says customer satisfaction has increased at his shop because end users are accessing their files faster via the cloud. "We've sped up access, as a lot of our users travel and remote access to our data center was slow," he says.
The same application speed-up can happen on a global basis, with business users having better access via a cloud provider's bigger bandwidth than they did to in-house servers. One strategy is to limit the application to updating itself once or twice each day, instead of hourly, for example, to keep performance issues to a minimum.