This week one of our applications is attempting to benchmark the performance of an Oracle database running on an Exadata machine versus running in the AWS cloud on an x1.16xlarge EC2. Our initial thought was to define the storage as two ASM disk groups, each consisting of four 400 gb disks at 6,000 IOPS each per ASM disk group. Assuming that a given data file was spread across all four disks in a disk group, we could have speeds approaching 20,000 IOPS.
During testing, Amazon recommended that the ASM data be placed on single disks running at 20,000 IOPS. The current maximum IOPS that can be provisioned is 20,000 IOPS. Additionally, AWS requires that the IOPS:gb ratio be maintained at a maximum of 50:1. In order to meet these requirements, we decided to create two mount points, each of 1000 gb at 20,000 IOPS. An additional 1000 gb at 20,000 IOPS was also added for the archive logs.
We created a json template to spin up the three disks and mounted them to the existing EC2. Before mounting, make sure that the mount points are not already being used by currently attached disks. After the disks are mounted, we attached them to ASM and waited for the rebalance to complete. After the rebalance was completed, we dropped the old disks that were provisioned at 6000 IOPS. The application saw a 67% decrease in elapsed time on some of the queries.
