Amazon EC2

• Elastic Compute Cloud = Infrastructure as a Service
• Rent virtual machines (EC2 instance)
• Store data on virtual drvies (EBS)
• Distrubute load accross machines (ELB)
• Scale the services using an auto-scaling group (ASG)

EC2 sizing & configuration

• OS: Linux, Windows, Mac OS
• Compute power & cores (CPU)
• How much RAM
• How much store space
  - Network attached (EBS & EFS)
  • Hardware attached (EC2 Instance Store)
• Network card: speed, Public IP
• Firewall rules: security group
• Boostrap script: EC2 User Data

User Data

• Bootstrap instance using EC2 User Data script
• Bootstrapping is launching commands when a machine starts
• Script is ran once at the instance first start
• Used to automate boot tasks such as
  - Install updates, software
  • Downloading common files
  • etc
• Runs with root user = has sudo rights

Instance types [1]

• vCPU: number of cores
• NVMe SSD: attached to EC2 instance
• Choose instance that best fits our application and use on demand

• Create key pair for SSH utility to access EC2 instance

• .pem: for linux, mac, windows >=10
• .ppk: windows <10

• The script get executed when a machine starts
• Update -> install httpd webserver -> write a html file for the webserver

• Instance Id: unique identifier for the instance
• Public IPv4 address: Address to access EC2 instance
• Private IPv4 address: Address to access EC2 instance internally on the AWS network

Restart the instance!

• Public IPv4 address has been changed to 3.250.26.200
• Private IPv4 address does not change

Instance types [2]

Instance name has naming convention ex) m5.2xlarge

• m: instance class
• 5: generation
• 2xlarge: size within the instance class

There are different types of EC2 instances optimised for different use cases

General Purpose

For diversity of workloads such as web servers or code repo

• Balance between compute, memory, and networking

• T, M family

Compute Optimized

For compute-intensive tasks that require high performance processors

• Batch processing workloads
• Media transcoding
• High performance web servers
• High performance computing (HPC)
• Scientific modeling & machine learning
• Dedicated gaming servers

• C family

Memory Optimized

For processing large data sets in memory

• High performance, relational/non relational databases
• Distributed web scale cache stores
• In memory database for Business Intelligence
• Real time processing of big unstructured data

• R, X1, High Memory, z familty

Storage Optimized

For storage intensive tasks that require high, sequential read and write access to large data sets on local storage

• High frequency online transaction processing (OLTP)
• Relvational & NoSql databases
• Cache for in-memory databases (ex. Redis)
• Data warehousing app
• Distributed file systems

Purchasing Options

On-Demand

• Pay for what you use
• Linux or Windows: billing per second after the first minute
• Others: billing per hour
• Highest cost but no upfront payment
• No long term commitment

For short term and un-interrupted workloads where you can't predict how the applicatoin will behave

Reserved

• Up to 72% discount compared to On-demand
• Reservation period: 1 year or 3 years (More discount for longer reservation)
• Payment options: no upfront, partial upfront, all upfront (More discount =>)
• Scope: Regional or Zonal (reserve capacity in an AZ)
• Buy and sell in the Reserved Instance Marketplace

For steady-state usage applications like database

• Convertible Reserved Instance
  - Change type, familty ,os scope, and tenancy -> more flexibility
  - Up to 66% discount

• Reserve a specific instance attributes (Instance Type, Region, Tenancy, OS)
  - Reserved Instances - long workloads

  • Convertible Reserved Instances - long workloads with flexible instances

Saving Plans

• Get a discount based on long term usage (up to 72%)
• Commit to a certain type of usage ($10/hour for 1 or 3 years)
• Usage beyond Savings Plans is billed On-Demand
• Locked to a specific instance familiy & AWS region
• Flexible across
  - Instance size
  • OS
  • Tenancy

Spot Instances

• The most cost-efficient
• Discount up to 90% compared to On-demand
• Can lose it at any point if max price is less than the current spot price (ex. bidding)
• Less reliable

For workloads that are resilient to failure like batch jobs, image processing, any distributed workloads, workloads with flexible start and end time

• Not for critical jobs or database

Dedicated Hosts

• The most expensive
• A physical server fully dedicated to your use
• Compliance requirements and existing server bound software liences
• Purchasing options
  - On-demands: pay per second for active Dedicated Host
  • Reserved: 1 or 3 years

For software that has complicated licensive model or strong regulatory or compliance needs

Dedicated Instances

• Instances run on hardware that's dedicated to you
• Can share hardware with other instances in same account
• No contorl over the instance placement

• Dedicated instances: own instance in own hardware
• Dedicated hosts: access to a physical server and gives visibility to the lower level hardware

Capacity Reservations

• Reserve On-Demand instance in a specific AZ for any duration
• Always have access to EC2 capacity when you need
• No time commitment, No billing discounts
• Combine with Regional Reserved Instances and Saving Plans to benefit from billing discounts
• Charged on demand whether you run instance or not

For short term, uninterrupted workloads that needs to be in a specific AZ

Spot Instances

• Define max spot price and get the instance while current spot price < max
• When price > max, choose stop or terminate with a 2 minute grace period

• Much cheaper than On-Demand price

Spot request

• Desired number of instances
• Maximum price
• Request type: One-time | Persistent
  - One-time: If spot request is fullfilled, the spot request will go away
  • Persistent: Even though spot instances are stopped, the spot request will automatically restart the instances
• Valid from, Valid until: Start and expiration
• Launch specification: ex) instance type, os, az..

Termination

• Can only cancel Spot Instance requests that are open, active, or disabled
• Cancelling a Spot Request does not terminate instances
• First cancel a Sport Request and then terminate the assoicated Spot Instances

Spot Fleets

• Set of Spot Instances + optional On-Demand Instances
• Define possible launch pools (instance type, os, az..)
• Fleet chooses the best one from the multiple launch pools
• Fleet stops launching instances when reaching capacity or max cost

Strategies to allocate

• Lowest price from the pool: cost optimization, short workload
• Diversified: distributed across all pools, great for availability and long workloads
• Capacity Optimized: pool with optimal capacity for the number of instances
• price Capacity Optimized: pools with highest capacity and then select the one with lowest price (best choice for most workloads)

Spoot Fleets chooses the best that suits one of the 4 options from launch pools