Design Cost Optimized Architectures

How often and how quickly do you need to access your data?
How large is your data set?
How transient is your data?
Clickstream
How much are you prepared to pay to store the data?
Durability: The average, expected annual data loss
S3 Standard, S3 IA (infrequent access), S3 Archive
Does your data require high IOPS?
EBS block storage
Throughput - HDD
IOPS - SSD
- Remember the "provisioned iops" ssd has much higher IOPS... 32000 iops
- cannot be boot volume
- probably need to pay monthly or annual instead of by minute/hour
S3 Resiliency: Almost all S3 Resilient against single AD (except for 1-zone IA)
Throughput oriented, lowest cost: Cold HDD (sc1)
Exam Question: I have an on demand instance, it costs $10/hour, I started at 930 AM and terminated at 1100 AM.
You will get charged for 2 hours (not 1.5 hours)
Some instances you can pay per second

Cheapest option, up to 90% off the price
You place a bid on amazon EC2 instances, you get unused extra capacity.
Instance can get terminated at any time.
You get a 2 minute warning, so there's solutions.
Apparently we can predict this and shut down the application.
Also apparently you don't get kicked off these very often most of the time, but you COULD BE!!

But instructor says this is changing, so all users will have the same AZ in the future
Users are binned because everyone would use the first one: us-east-1a

Up to 75% discount compared to on-demand instances
Capacity Reservation - 1 to 3 years
AWS particular AZ can run out of capacity on-demand for a particular type of instance
RI Marketplace