markrubens - Fotolia
Unless you work for a startup born in the cloud or a company prepared to go instantly "all in" with Amazon Web Services, chances are you'll manage a hybrid cloud infrastructure over the next few years.
And for some IT shops, cost analyses of large installations of traditional enterprise applications don't work out in favor of hosting them in the Amazon Web Services (AWS) public cloud for the foreseeable future.
A visual communications company based in the Pacific Northwest is one such shop -- it has a mix of infrastructures that includes 400 VMware hosts running 6,000 VMs, another 1,600 physical hosts, 300 instances running in an OpenStack private cloud and a handful of applications stood up by developers in AWS.
The company is evaluated based on EBITDA, which favors capital investments over ongoing operation expenses, said a senior director of platform engineering for the company, who requested anonymity.
"I would prefer to run apps in AWS because it allows the offloading of the operational responsibility of the infrastructure to someone who has much more scale than I do," he said. "However, the method of paying for subscription services like this and their impact on EBITDA make it a less favorable platform in an apples-to-apples comparison."
Meanwhile, there are certain capabilities AWS can always deliver that are cost prohibitive, too complex for the company to deliver or aren't prioritized on the company's roadmap -- for instance, localized delivery of services, such as Simple Storage Service, the accommodation of large swings in capacity need and services that are delivered via platform as a service. Thus, this shop won't move out of AWS entirely, either.
FlightStats Inc., a global data service company in the aviation space located in Portland, Ore., relies on an AWS hybrid cloud to pass the best price for data along to customers.
"We acquire data, crunch it and get it ready for consumption in our own private cloud data centers, [because] we get a better dollar-per-performance in our own data center," said Alex Witherspoon, vice president of platform engineering for FlightStats.
The key is to avoid running instances in AWS all the time, Witherspoon said.
"Where we see people struggle is when they put workloads that don't fit AWS' model … they end up paying for it, quite literally, in dollars," Witherspoon said.
Many shops -- especially in the enterprise -- take large, heavy, single-server workloads from old-world IT and try to run them in AWS, Witherspoon said.
"These things are literally hundreds or thousands of dollars to run per month for a single instance, because they're running these big, massive, old school databases … things that require lots and lots of memory," Witherspoon said. "It's when you need 100 GB of memory in Amazon that you've really gone way over it."
Organizations constantly evaluate the return on investment throughout a project, according to Matthew Scott, vice president of public cloud for Datapipe Inc., a provider of managed hosting services for AWS based in Jersey City, N.J.
"You look at some of these legacy applications … does it really make sense for an organization to spend several million dollars to retrofit an application that was never designed to run in cloud?" Scott said. "That's a very early conversation we have with a lot of large organizations."
Some are willing to make the investment, Scott said, but others wonder, why refactor an application when they have amortization still recognized on the hardware investments they made?
This is especially questionable when contractual obligations are in place for on-premises hardware use.
Christopher Smithmanager of infrastructure engineering for TaylorMade
"I have [client] organizations that have enormous hardware buys that are still in flight, and they are contractually going to be getting N number of servers for the next X years," said Patrick McClory, director of automation and DevOps for Datapipe. "For them to stop using those and move to cloud would actually be a hit to their stock price -- they'd have to write that down and it would be visible."
One Datapipe customer, TaylorMade-Adidas Golf Company, a manufacturer of golf equipment and apparel based in Carlsbad, Calif., uses AWS for disaster recovery on Oracle ERP systems, but has not seen the need to run production servers in AWS.
"[Public cloud] is cheap, but it's not that cheap," said Christopher Smith, manager of infrastructure engineering for TaylorMade. "If you start spinning up lots of storage and [have] lots of I/O in and out, it would easily surpass what we pay for our server room maintenance in a year … it's not easy just to say, 'This goes to the cloud because we're going to save loads of money.'"
The company has a backup of an Oracle ERP system in the cloud for disaster recovery purposes, but the rest of the company's infrastructure remains on premises. Smith said he hadn't ruled out going to the cloud for DR on more apps, but wasn't sure whether the company's Oracle database-heavy production apps would ever find a home on AWS.
AWS hybrid cloud future looks bright
Hybrid cloud challenges remain for AWS users
Get a better understanding of hybrid cloud workflow management