Since then, Entonos has completely rewritten the workflow so that
- Only AWS native services used (Univa’s schedulers are not needed)
- Avoids EBS storage and network traffic
- Self-orchestration of containers to significantly reduce traffic to/from S3
- Automatic selection of AWS region/AZ for lowest cost computing at submission
- Automation of highly optimized computational executable(s) creation
- Automation of custom AMI/Container/VPC/Batch setup and tear down
- Allow for legacy run scripts, including openMP/openACC and MPI simulations
This results in 20-30% cost reduction (equivalently ~97% of total cost is only for compute) and much more flexible workflows.
Interested? We would like to hear about your challenge.