Detailed cost analysis and cost optimization

[JiaweiZhuang]

Computing cost

Preliminary estimation of computing cost, with spot instances:

• 1-month global 4x5 will cost ~$1
• 1-month 0.25 nested NA will cost ~$15
• 1-year global 2x2.5 simulation will cost ~$50 (still covered by the $100 education credit).
• For even longer runs, recommend to apply for research grants!

TODO:

• Provide actual timing results and cost calculation. Timing tests should be saved as a snapshot to be reproducible by everyone.
• Compare with the hourly cost of Pleiades: https://www.hec.nasa.gov/user/policies/sbus.html. My preliminary calculation is that standard instances are more expensive but spot instances are actually cheaper.

Storage cost

• 100 GB of data will cost $10/month on EBS, cost $2.3/month on S3, $0.4/month on glacier, and $9 to transfer out of AWS.
• The cost grows linearly (pretty quickly) as the data size increases. While 100 GB costs almost nothing, 1 TB does lead to a significant billing if you don't manage it carefully. For really huge data, again recommend to apply for research grants.

TODO:

• Estimate the output data size for typical research projects.

Cost saving tricks:

• Data egress is expensive. More cost effective to keep data on AWS and perform data analysis there.
• Apply for Data Egress Discount to Researchers
• Only download key variables to local machines, if necessary.

[JiaweiZhuang]

Would be nice to have two tables. Once is the cost rate (i.e. hourly/monthly cost) for commonly used services. Another is the actual total cost for a sample project like in #4

[JiaweiZhuang]

Done, as the paper is submitted