In a migration to the cloud, there's a "bubble of expense" during the period when you're setting up an application in the cloud but still operating it on premises. "The migration bubble balloons costs," warned Ken Griffin, senior director of IT operations at Harvard Business Publishing, which made its own migration into Amazon Web Services.
Harvard Business Publishing is a fully-owned subsidiary of Harvard University. It publishes the Harvard Business Review, which along with its book publishing division, is one of HBP's business units. It and two other units, the Harvard Business School Executive Education and the Harvard Business School MBA program, migrated out of rented data center space and into Amazon Web Services at the end of 2015.
IT at HBP consists of Griffin and six operations people who work with the talent available in the business units to constantly maintain and update an education and publishing site. The site publishes 16-to-20 articles on a daily basis.
Griffin enlarged upon this point during a session, Cloud Migration: After the Honeymoon, at Interop ITX in Las Vegas last month. InformationWeek offered a preview of his talk on April 28, At Harvard Business Publishing, It Was a Do-or-Die Move to the Cloud. In his actual session, he hit upon some concrete conclusions about the cost of going to the cloud.
Want to learn more about the prominence of financial services at AWS Re:Invent? See FINRA Commits Mission-Critical App to Amazon Cloud.
Griffin urged the three business units of his 656-employee company to make the migration, even when they encountered the expense bubble. At one point, he gave them an ultimatum of several months to help IT finish the job or else. To reinforce the message, he posted his notice of the upcoming termination of rented space with HBP's data center provider. "If we missed the deadline, we'd be locked into another year of the data center contract," he noted. As business units hurried their applications into the cloud in late 2015, Griffin gave everyone a three-month breathing space. Then he began pressing for refactoring applications to take advantage of the services in AWS and lower their costs.
During the lead up to the data center shut down, "I told them not to worry about costs," he said. Three months later, "I said, 'Now that we're here, let's optimize our instances to reduce costs," he added.
The lift-and-shift phase of applications was necessary to get everyone moving without re-engineering their apps. Paying more to operate two sets of applications in the "bubble" was necessary to get to the gains and savings on the other side of the migration, he said. "We paid the 'silly' tax," he said.
Once in AWS, Griffin found some applications running 24 hours a day on large servers to meet peak loads, when much of the time they could operate at steady-state on a much smaller server. When the cost benefits of moving to automated load balancing and scaling were pointed out, business units began moving their servers to smaller instances managed by AWS Elastic Beanstalk. Elastic Beanstalk can provision the application at an appropriate server size for normal operation, then scale it up as needed.
Likewise, instead of provisioning 24X7 larger servers, AWS ElasticCache service loads frequently accessed content or data into a cache directly from S3 storage, allowing multi-media, content heavy Web sites to operate with lower latencies without needing bigger servers. These were the examples of the "low-hanging fruit" in cost savings, Griffin said.
As Harvard Business Publishing business units began to use Amazon's Relational Database Service, including Oracle as a service, Aurora and DynamoDB, they could decommission large standing database servers that had initiated to make their move.
"Cost wasn't a primary driver in moving to the cloud, but we're absolutely saving money. We couldn't get to the savings without refactoring applications," he noted. He estimated an 18% reduction in overall IT costs as a result of the move.
But those costs won't materialize until you can get out of the "bubble." Griffin warned his listeners at Interop ITX, "The migration bubble is like a nuclear reactor. The longer you stay in it, the more likely you'll die," he said.
To get the business units to better understand their options, he sent his six core IT staff members along with 16 employees from the business units to AWS's Re:Invent Show in Las Vegas last year to give them a taste of Amazon "Kool Aid."
The more important long term goal for Griffin is revamping the organization into a digital company, able to respond faster to challenges and fully engaged with keeping up with the pace of change in publishing. The core IT staff is also his DevOps team, and it's using Python Lambda functions in the cloud for their next generation applications.
With development in the cloud, "new apps are being developed with a cloud-native focus... A shift to microservices is underway and a 'disposable app' mindset is taking hold," where applications are not developed to last forever but developed to meet an immediate business need, then supplanted by another at a later date.
"If we had kept the data center, we wouldn't have the resources to do what we've already done. There's no going back. We've burned the ships," he said.