Skip to content

Latest commit

 

History

History
188 lines (182 loc) · 11.6 KB

Chapter 8 - Cost Management.md

File metadata and controls

188 lines (182 loc) · 11.6 KB

Cost Management Module

  • Concepts
    • Capital Expenses (CapEx)
      • Money spent on long-term assets like property, buildings and equipment etc
    • Operational Expenses (OpEx)
      • Money spent for on-going costs for running the business. Usually considered variable expenses
    • Total Cost of Ownership (TCO)
      • A compherhensive look at the entire cost model of a given decision or option, often including both hard costs and soft costs
    • Return on Investment (ROI)
      • The amount an entity can expect to receive back within a certain amount of time given an investment
    • TCO and ROI
      • Many times, organizations don't have a good handle on their full on-prem data ceontre costs (power, cooling, fire suppression etc)
      • Soft costs are rarely tracked or even understood as a tangible expense
      • Learning curve will be very different from person to person
      • Business plans usually include many assumptions which in turn require support organizations to create derivative assumptions - sometimes layers deep
  • Cost Optimization Strategies

    • Appropriate Provisioning

      • Provision the resources needed and nothing more
      • Consolidate where possible for greater density and lower complexity (multi-database RDS, containers)
      • CloudWatch can help by monitoring utilization

    • Right-Sizing

      • Using lowest-cost resource that still meets the technical specifications
      • Architecting for most consistent use of resources is best versus spikes and valleys
      • Loosely coupled architectures using SNS, SQS, Lambda and DynamoDB can smooth demand and create more predictability and consistency

    • Purchase Options

      • For permanent applications or needs, Reserved Instances provide the best cost advantage
      • Spot instances are best for temporary horizontal scaling
      • EC2 Fleet lets users define target mix of On-Demand, Reserved and Spot Instances
      Instance OnDemand Reserved 1 Year All Upfront Spot Instance
      m5.2xlarge 0.384 0.229 (40% less) 0.0798 (79% less)

      • Geographical Selection

        • AWS # can vary from region to region
        • Consider potential savings by locating resources in a remote region if local access is not required
        • Route 53 and CLoudFront can be used to reduce potential latency of a remote region
        Region S3 Standard Storage First 50 TB
        us-west-2 0.023/GB
        us-west-1 0.026/GB
        ap-northeast-1 0.025/GB
        sa-east-1 0.0405 per GB

      • Managed Services

        • Leverage managed services such as MySQL RDS over self-managed options such as MySQL on EC2
        • Cost savings gained through lower complexity and manual intervention
        • RDS, RedShift, Fargate and EMR are great examples of fully-managed services that replace traditionally complex and difficult installations with push-button ease.

      • Optimized Data Transfer

        • Data-in to AWS, no cost
        • Data going out and between AWS regions can become significant cost component
        • Direct Connect can be more cost-effective option given data volume and speed
  • AWS Tagging
    • The Number one best thing that can be used to manage the AWS assets
    • Tags are just arbitrary name/value pairs that you can assign to virtually all AWS assets to serve as metadata
    • Tagging strategies can be used for Cost Allocation, Security, Automation, and many other uses
      • Example: A tag can be used in an IAM policy to implement access controls to certain resources
    • Enforcing standardized tagging can be done via AWS Config Rules or custom scripts
      • Example, EC2 instances not properly tagged are stopped or terminated
    • Most resources can have up to 50 Tags
  • AWS Resources Groups
    • Resource Groups are groupings of AWS assets defined by tags
    • Create custom consoles to consolidate metrics, alarms and config details around given tags
    • Common Resources Groupings
      • Environments (DEV, QA, PRD)
      • Project Resources
      • Collection of resources supporting key business processes
      • Resources allocated to various departments or cost centres
  • Spot Instances and Reserved Instances
    • Reserved Instances

      • Purchase (or agree to purchase) usage of EC2 instances in advance for a significant discount over On-Demand #

      • Provides capacity reservation when used in a specific AZ

      • AWS Billing automatically applies discounted rates when an instances is launched that matches purchased RI

      • EC2 has three RI types

        • Standard
        • Convertible
        • Scheduled

      • Can be shared across multiple accounts within Consolidated Billing

      • If RIs are not needed, then it can be sold on the Reserved Instance Marketplace

      • Standard vs Convertible

        Standard Convertible
        Terms 1 year, 3 year 1 year, 3 year
        Average Discount off On-Demand 40% - 60% 31% - 54%
        Change AZ, Instance Size, Networking Type Yes via ModifyReservedInstance API or console Yes via ExchangeReservedAPI or console
        Change instance family, OS, Tenancy, Payment Options No Yes
        Benefit from Price Reduction No Yes
        Sellable on Reserved Instances Marketplace Yes (Sale proceeds must be deposited in a US bank account) Coming soon

      • RI Attributes

        • Instance Type - designates CPU, memory, networking capability
        • Platform - Linux, SUSE Linux, RHEL, Microsoft Windows, Microsoft SQL Server
        • Tenancy - Default (shared) tenancy or Dedicated tenancy
        • Availability Zone (Optional)
          • If AZ is selected, RI is reserved and discount applies to that AZ (Zonal RI).
          • If no AZ is specified, no reservation is created by the discount is applied to any instance in the family in any AZ in the region (Regional RI)
          • Zonal RI can be changed to Regional AI or vice versa via the console or ModifyReservedInstance API
    • Spot Instance
      • Excess EC2 capacity that AWS tries to sell on market exchange basis
      • Customer creates a Spot Request and specifies AMI, desired instance types and other key information
      • Customer defines highest price willing to pay for instance. If capacity is constrained and other are willing to pay more, the instance can be terminated or stopped
      • Spot request Types
        • Fill and Kill (One Time only)
          • Spins up a spot instance if the price is higher than the going rate
          • It will get rid of that request
          • When the instance is terminated ephemeral data is lost
        • Maintain
          • Will re provision a spot instance if by chance the price rises too high and it gets terminated.
          • When the price goes back down maintain will recreate another spot instance
          • Same applies if the instance is manually terminated
          • Instances can be configured to Terminated, Stop or Hibernate until price point can be met again
        • Duration
          • A finite period time can be specified
            • For example, a spot instance can be booked for 3 hours
            • After 3 hours, the instance is terminated and the request is removed alt text
    • Dedicated instances vs Dedicated Hosts
      • Dedicated instances
        • Virtualized instances on dedicated hardware
        • May share hardware with other non-dedicated instances in the same account
        • Available as On-Demand, Reserved Instances and Spot Instances
        • Cost additional $2 per hour per region
      • Dedicated Hosts
        • Physical servers dedicated a single customer
        • Complete control of which instances are deployed on that host
        • Available as On-Demand or with Dedicated Host Reservation
        • Useful if there is a server-bound software licenses that use metrics like per-core, per-socket or per-VM
        • Each dedicated host can only run one EC2 instances size and type
  • Cost Management Tools
    • AWS Budgets
      • Allows to set pre-defined limits and notifications if nearing a budget or exceeding the budget
      • Can be based on Cost, Usage, Reserved Instance Utilization or Reserved Instance Coverage
      • Use as a method to distribute cost and usage awareness and responsibility to platform users
    • Consolidated billing
      • Enable a single payer account that's locked down to only those who need access
      • Economies of scale by bringing together resource consumption across accounts
    • Trust Advisor
      • Runs a series of checks on the resources and proposes suggested improvements
      • Can help recommend cost optimization adjustments like reserved instances or scaling adjustments
      • Core checks are available to all customers
      • Full Trusted advisor benefits require a business or enterprise support plan
  • Exam Tips
  • Pro Tips
    • Be extra careful around the TCO and ROI minefield.
      • Take advise from a finance person
      • Find ways to translate soft costs into hard costs. Refer to academic research or studies
      • Harvard Business Review and MIT Sloan Management Review have articles on Organizational Change Management quantifying soft costs.
      • Take care when looking at the case studies from a vendor -
    • The real benefit of a Cloud Migration is in Agility and Flexibility. Cost alone is not the strongest business case.
      • Initial migration costs are going to raise because it is something different
      • Cost is a malleable, it can go up and down
      • The real benefit is the agility and flexibility. It translates to cost savings, customer happiness and ability to respond to external or internal threats. They also translate into innovation.
    • Think of cost optimization as a long-term effort - don't spend too much time on trying to micro-manage it up-front
    • Implement a tagging strategy out of the gate!
    • Be aggressive in formulating a pilot project: Large-scale benefits make for dramatic business cases, where small-scale winds can easily be ignored