Cloud Computing has become the platform of choice for the majority of IT organizations. The mantra “cloud first” is chanted throughout both the private and public sectors. According to Gartner, “cloud only” will be the defacto policy for all new IT initiatives at more than 30% of large enterprises by 2020. Gartner also ranks Amazon Web Services (AWS) as the dominant player in public cloud Infrastructure as a Service (IaaS), having captured 44 percent of the market as early as 2016. To put that into perspective, the next biggest player, Microsoft Azure, has 7 percent market share. Finally, Gartner expects more than half of global enterprises exploiting the cloud today will adopt an all-cloud strategy by 2021.
If your organization runs a Unisys mainframe today, you may be wondering if your tried-and-true core applications will be left behind. And even though you understand that migrating mainframe applications to open-systems and the cloud is possible, you may still have a healthy dose of skepticism as to whether AWS has the power to meet the demands of your mainframe applications. We had the same question here at Astadia, so we decided to find out. We took one of our own Unisys mainframe applications, migrated it to AWS, and conducted several months of performance testing. Our results were not quite what we expected. They were better.
We submitted workload for 200 users every 2 minutes for a total of 1000 users. Each user submitted 700 transactions for a total of 700,000 transactions per testing scenario. When conducting these tests on a single, mid-level m4.4xlarge Amazon Elastic Cloud Compute (EC2) instance, we routinely achieved 1600 transactions per second and sub-second response times. But, even more interesting and exciting were the results when testing a load-balanced scenario of multiple EC2 instances.
We reconfigured our AWS environment to run the mainframe application using 8 instances of EC2 type m4.large (the lowest powered m4 type offered), and used the AWS Elastic Load Balancer to distribute the workload evenly among the application instances. We also doubled the workload, simulating 200 users every 2 minutes until we reach 2000 users, each submitting 700 transactions for a total of 1,400,000 transactions. Under this configuration, we easily achieved 2200 transactions per second at sub-second response times, taking just 18 minutes to complete the test. Remember, this was the lowest powered m4 instance type with just 2 CPUs and 8GB RAM per instance.
These results lead us to conclude that AWS is indeed capable of handling mainframe workload. When combined with its reliability, flexibility, and low cost, AWS is the logical choice as a strategic platform for Unisys mainframe applications.