Branden Moore from NAG (formerly the Numerical Algorithms Group) recently gave a talk to the Society of HPC Professionals on migrating a seismic workload to the cloud. He subsequently started a blog on the approach which we summarize here.
Moore defines a ‘cost of solution’ (CoS) metric as the overall cloud costs incurred by running the workload, including data transfer and ancillary cloud services required. The analysis considers a NAG-developed reverse time migration (RTM) benchmark, comparing a ‘lift and shift’ approach, porting the NAG code to an 8 petabyte Amazon FSx for Lustre managed service. Here a 10,000 shots RTM would run for 47 days and cost ‘just shy of $4 million’.
An alternative approach using Amazon’s ‘i3en’ virtual machines with large local storage with one instance per shot. Rejigging the code to the VMs lowered the CoS to $1.64 million but increased run time to 52 days.
Finally, Moore figured the costs of using the new AWS Graviton 2 processor which perform HPC tasks well and cost less than other x86 processors. They were combined with the previous i3en instances to arrive at a CoS of only $1.16 million and a much-improved run time of 21 days. Adding extra cloud resources actually reduced the total cost.
More from the new NAG Cloud HPC Migration Service.
© Oil IT Journal - all rights reserved.