Title

AWS re:Invent 2023 - Capacity, availability, cost efficiency: Pick three (CMP207)

Summary

The session focused on balancing capacity, availability, and cost efficiency in AWS.
AWS offers a pay-as-you-go pricing model and capacity reservations to ensure operational resiliency.
AWS infrastructure includes 102 availability zones across 32 geographic regions.
EC2 instance characteristics are crucial for selecting the right instance for workloads.
AWS provides multiple pricing models: On-Demand, Savings Plans, and Spot Instances.
Flexibility is key in managing capacity and availability, and AWS offers tools like auto scaling groups, EC2 fleet, and attribute-based instance selection.
Savings Plans offer up to 72% savings and come in different types, with Compute Savings Plans being the most flexible.
Capacity management involves planning, acquiring, monitoring, and optimizing resources.
Salesforce shared their experience with capacity planning on AWS, highlighting the use of Spot Instances, regional flexibility, and instance type diversity.
Amazon EC2 Capacity Blocks is a new purchasing option for reserving GPU capacity for ML workloads on a per-job basis.

Flexibility is central to AWS's approach to capacity management, allowing customers to adapt to changing needs and optimize costs.
AWS's robust infrastructure with numerous availability zones and regions provides significant advantages in terms of capacity and availability.
Understanding EC2 instance naming conventions can help customers choose the most effective instances for their workloads.
Combining different AWS pricing models can lead to significant cost savings and operational efficiency.
Tools like attribute-based instance selection and Spot Placement Score can greatly simplify capacity management and planning.
Salesforce's case study illustrates the practical application of AWS's capacity management strategies, emphasizing the importance of instance diversity and regional flexibility.
Amazon EC2 Capacity Blocks offer a new way to manage GPU capacity for machine learning projects, providing flexibility and cost control for short-term, high-intensity workloads.
The session highlighted the importance of continuous learning and staying updated with AWS's evolving services and tools to optimize compute resources effectively.