Utopia Migrates SAP Systems to AWS and Uses Machine Learning

Utopia conducted a six-month evaluation of cloud service providers, comparing Amazon Web Services (AWS), Google Cloud Platform, and Microsoft Azure. It concluded that cloud migration would be 15 percent faster on AWS due to the maturity of the platform, easily configurable builds, and automation capabilities. “AWS environments are easily scalable, convertible, and highly fault-tolerant, matching our needs of achieving high business agility with minimal disruptions,” Ganjiwale explains.

Furthermore, the overall cost of AWS was estimated to be 20 percent lower than Microsoft Azure, including training and credits offered by AWS. Another aspect in favor of AWS was the range of instance types available. To get the same price point with Microsoft Azure, the business would have needed bigger instances, which would have required it to change the way servers were structured. “With AWS, we fit our existing instances in without needing to resize them, which was important because we wanted to migrate as quickly as possible. The newer AWS instance sizes have faster processing capabilities, and the pace at which AWS releases new instance styles fits our needs,” Ganjiwale adds.

Utopia now uses Amazon Elastic Cloud Compute (Amazon EC2) R5 instances and Amazon EC2 M5 instances, which are both well suited to memory-intensive SAP HANA workloads, while Amazon EC2 P3 instances are used for machine learning (ML).

To control costs, Utopia is taking advantage of Amazon EC2 Spot Instances. The company has ramped up its ML efforts in the past two years, forming a 20-person team dedicated to developing ML capabilities. Amazon EC2 Spot Instances have been ideal for training ML models, which require a huge amount of memory power to process millions of records at once. In the pilot stage, conversion to Spot versus On-Demand Instances for ML model training resulted in 13 percent cost savings for the company. “We expect to save even more going forward once we understand how to properly optimize Amazon EC2 Spot Instances,” says Bipin Ramanujan, vice president at Utopia.

Utopia relies on AWS Virtual Private Network (AWS VPN) tunneling as a secure way to send data between its clients’ cloud or on-premises networks and SAP systems. It also uses Amazon Elastic Kubernetes Service (Amazon EKS) for orchestrating a container-based SAP application for data integration. Utopia is now exploring Amazon SageMaker and Amazon Rekognition to automate the cleansing, standardization, and enrichment of data it has been doing manually for 16 years. “With machine learning on AWS, we’ve brought a new capacity for volume and accuracy we couldn’t achieve before,” says Peter Aynsley-Hartwell, CTO at Utopia. “It has enabled us to scale services that were predominantly people-driven to those that are now IP-driven.” The company hopes to generate new revenue streams by integrating ML into its MDG implementation with a pay-as-you-go subscription service.

For its SAP migration, Utopia moved 120 servers to the AWS Cloud in just three months, hitting its stretch target for the project; the original target was four months. Its strategy included a detailed proof of concept backed by careful risk management planning. During the migration, AWS solutions architects conducted intensive sessions on any issues that arose. “The AWS team was like an extension of our own,” says Ramanujan. “They worked closely with us to understand our requirements and designed a well-architected framework with different Availability Zones—helping us choose the best region for our business with the lowest cost.”

The savings since migration have been substantial, and Utopia can now handle more workloads at a lower cost. The business has reduced its monthly spend by 47 percent since moving from CenturyLink to AWS, while increasing its workloads by 40 percent with new ML projects. Aynsley-Hartwell credits his team, who carefully monitor their usage of servers and turned them on only when needed—a feature unavailable in the colocation data center environment. They have also saved money by switching to Amazon EC2 Convertible Reserved Instances for regular workloads such as data sanitization, in lieu of Standard Reserved Instances.

Utopia’s management is also pleased with the time savings gained on the AWS Cloud. For instance, when a client requested a trial of Utopia software before, the team would take one to two weeks to set up a dedicated test system. With AWS, this now takes just two days. “Being able to offer our clients a faster turnaround helps us immensely in selling software,” Aynsley-Hartwell says. “We can also encourage clients to move to the cloud if we demonstrate these benefits during the presales cycle.”

Looking ahead, as the company works to integrate more machine learning into its products, Aynsley-Hartwell is confident that it is on the right path. “We have a very good relationship with AWS. There’s a strategic alignment and we’ve learned a lot as well.”