Compute and Container

EC2

• EC2 = Elastic Compute Cloud

Instance Type

• AWS has the following naming convention m5.2xlarge`
  • m is the instance type
    • General Purpose - T, M, A, Mac
    • Compute Optimized - C
    • Memory Optimized - R, X, z, High memory
    • Storage Optimized - I, D, H
  • 5 is the generation (AWS improves theme over time)
  • 2xlarge is the size
• AWS Instance Types

User-Data

• That script is only run once at the instance the first start
• The EC2 User Data Script runs with the root user
• EC2 user data is used to automate boot tasks such as
  • Installing updates
  • Installing software
  • Downloading common files from the internet
  • Anything you can think of

Security Group

• All inbound traffic is blocked by default
• All outbound traffic is authorized by default
• If your application is not accessible by timeout then it's a security group issue
• If it gives a connection refused error then it's an application error or it's not launched
• Can reference to another security group

EC2 Instance Connect

• Connect to your EC2 instance within your browser
• Works only out-of-the-box with Amazon Linux 2
• Need to make sure port 22 is still open!
• No need to use your key file that was downloaded
• The “magic” is that a temporary key is uploaded onto EC2 by AWS

Purchasing Options

• On-Demand Instances
  • Highest cost but no upfront payment
  • Billing per hour
• Reserved Instances
  • up to 72% discount compared to on-demand
  • Reserve specific instance attributes (Instance type, Region, Tenancy, OS)
  • Reservation period: 1 year (+discount), or 3 years (+++discount)
  • Payment option: No upfront (+), Partial Upfront (++), All upfront (+++)
  • Scope: Regional or Zonal
  • Can buy or sell in Reserved Instance Marketplace
  • Convertible Reserved Instances
    • Can change the EC2 instance type, instance family, OS, scope, and tenancy
    • Up to 66% discount
• Savings Plans
  • Commit to eligible usage at a dollar-per-hour rate; uncovered usage is billed at the On-Demand price
  • Compute Savings Plans cover EC2 across instance families and regions, plus Fargate and Lambda
  • EC2 Instance Savings Plans are tied to one instance family in one AWS region; savings can reach 72%
  • Other plan types include SageMaker AI Savings Plans and Database Savings Plans
  • AWS Savings Plans types (checked 2026-09-30)
  • Proportional RI and Savings Plans Chargeback explains allocating centrally purchased RI/SP savings through AWS Billing Conductor
• Spot Instances
  • up to 90% discount (Most cost-efficient instances)
  • Instances can lose at any point in time if your max price is less than the current spot price
• Dedicated Hosts (Book an entire physical server, Control instance placement)
  • Physical server with EC2 instance capacity fully dedicated
  • Allow for the compliance requirements and licenses (BYOL)
  • Purchasing options: On-demand (pay-per-second) and Reserved
  • Most expensive option
• Dedicated Instances (no other customers will share your hardware)
  • Instances run on hardware that's dedicated to you
  • May share hardware with other instances in the same account
  • No control over instance placement (Can move hardware after Stop / Start)
• Capacity Reservations (reserve capacity in a specific AZ for any duration)
  • Reserve on-demand instances capacity in a specific AZ for any duration
  • No time commitment (Create/Cancel anytime), No billing discount

Spot Instance

• define max spot price and get the instance while the current spot price < max
• You can choose to Stop or Terminate in 2 minutes grace period
  
• You can only cancel Spot Instance requests that are open, active, or disabled
• Cancelling Spot Request does not terminate the instance

Spot Fleets

• Set of Spot Instance + (optional) On-Demand Instance
• The Spot Fleet will try to meet the target capacity with price constraints
  • Define possible launch pools (instance type, OS, zone)
  • Can have multiple launch pools, so the fleet can choose
  • Spot Fleet stops launching instances when reaching capacity or max cost
• Strategies to allocate Spot Instances
  • lowestPrice: from the pool with the lowest price (cost-optimized, short workload)
  • diversified: distributed across all pools (great for availability, long workloads)
  • capacityOptimized: pool with the optimal capacity for the number of instances
  • priceCapacityOptimized (recommended): pools with the highest capacity available, then select the pool with the lowest price (best choice for most workloads)
• Spot Fleets allow us to automatically request Spot Instances with the lowest price

Placement Groups

• Strategies
  • Cluster: Clusters instances into a low-latency group in a single AZ
  • Spread: spreads instances across underlying hardware (max 7 instances per group per AZ, Critical applications)
  • Partition: Spreads instances across many different partitions (which rely on different sets of racks) within an AZ, Scale to 100s of EC2 instances per group (Hadoop, Cassandra, Kafka)

Hibernate

• Support instance families: C3, C4, C5, I3, M3, M4, R3, R4, T2, T3, and more
• Root volume must be EBS and encrypted
• Can't be hibernated for more than 60 days
• Not support for bare metal instances
• Ram must be less than 150GB

AMI

• Amazon Machine Image is a customization of an EC2 instance image
• Built for a specific region (can copy to another region)
  

Load Balancer

• Spread the traffic across multiple servers downstream
• Expose a single point of access (DNS) to your application
• Seamlessly handle failures of downstream instances
• SSL termination (Can manage certificates using ACM)
  • Support SNI (Server Name Indication) to solve the problem of loading multiple SSL onto one web server (To serve multiple websites)
    • It's a newer protocol and requires the client to indicate the hostname of the target server in the initial SSL handshake
    • The server will then find the correct certificate or return the default one
    • Only work for ALB & NLB (Newer generation), CloudFront
• Sticky Sessions (Session affinity)
  • The same client always redirected to the same instance behind a load balancer
  • Works for CLB, ALB, and NLB
  • The cookie used for stickiness has an expiration date you control
  • Enabling stickiness may bring imbalance to the load over the backend
  • Cookie Names
    • Application-based Cookies
      • Custom cookie
        • Generated by the target
        • Can include any custom attributes required by the application
        • Cookie name must be specified individually for each target group
        • Don't use AWSALB, AWSALBAPP, or AWSALBTG (reserved for ELB)
      • Application cookie
        • Generated by the LB
        • The cookie name is AWSALBAPP
    • Duration-based Cookies
      • The cookie generated by the LB
      • Cookie name is AWSALB for ALB, AWSELB for CLB
• Cross Zone Load Balancing
  • ALB
    • Enabled by default (Can disable at the target group)
    • No charges for inter-AZ data
  • NLB & GWLB
    • Disabled by default
    • Pay charges (&) for inter-AZ data if enabled
  • CLB
    • Disabled by default
    • No charges for inter-AZ data if enabled
      
• Connection Draining / Deregistration Delay
  • Time to complete "in-flight requests" while the instance is de-registering or unhealthy
  • Stops sending new requests to the EC2 instance which is de-registering
  • Between 1 to 3600 seconds (default 300s)
  • Can be disabled (set value to 0)
  • Set to low value if your requests are short
• LB Type
  • (Deprecated) Classic Load Balancer (CLB) (HTTP, HTTPS, TCP, SSL)
  • Application Load Balancer (ALB) - layer 7 (HTTP, HTTPS, WebSocket)
    • Routing tables to different target groups (path, hostname, header, or query string)
    • Have a port mapping feature to redirect to a dynamic port in ECS
    • Target groups: ALB can route to multiple target groups
      • EC2
      • ECS tasks
      • Lambda functions
      • IP Address
    • Health checks at the target group level
  • Network Load Balancer (NLB) - layer 4 (TCP, UDP, TSL)
    • Less latency ~100ms (vs 400ms for ALB)
    • One static IP per AZ and support assign EIP
    • Use for extreme performance
    • Not included in the free tier
    • Target groups:
      • EC2
      • IP Addresses - must be private IPs
      • ALB
    • Health checks support the TCP, HTTP, and HTTPS protocols
  • Gateway Load Balancer (GWLB) - layer 3 (IP Protocol)
    • Deploy, scale, and manage a fleet of third-party network virtual appliances in AWS
    • Examples: Firewalls, Intrusion Detection and Prevention Systems, Deep Packet Inspection Systems, payload manipulation, etc
    • Combines the following functions
      • Transparent Network Gateway
      • Load Balancer
    • Uses the GENEVE protocol on port 6081
    • Target groups:
      • EC2
      • IP Addresses - Must be private IPs

Autoscaling Group (ASG)

• Manage the number of EC2 Instances depends on the workload
• Automatically register new instances to a load balancer
• Re-create an EC2 instance in case a previous one is terminated
• FREE!
• Strategies
  • Dynamic
    • Target Tracking Scaling: Want the average CPU to stay at around 40%
    • Simple / Step Scaling: When the CloudWatch alarm is triggered, then add 2 units
    • Scheduled Actions: Increase the min capacity to 10 at 5 PM on Fridays
  • Predictive
• Good metrics
  • CPU utilization
  • RequestCountPerTarget
  • Average Network In / Out
• Scaling Cooldowns
  • After a scaling activity happens, You are in the cooldown period (default 300 seconds)
  • During the cooldown period, the ASG will not launch or terminate additional instances (Wait for stabilize)

Launch Template

• Attributes
  • AMI, Instance type
  • EC2 User Data
  • EBS Volumes
  • Security Group
  • SSH Key pair
  • IAM Role for EC2 instances
  • Network + Subnets information
  • LB information
• Min/Max/Desired Capacity
• Scaling Policy
  • Based on CloudWatch alarms (metrics such as Average CPU)

Launch Configuration

• Deprecated

ECS

• Elastic Container Service
• Launch ECS Tasks on ECS Clusters
• Launch Type
  • EC2
    • Each EC2 Instance must run the ECS Agent to register in the ECS Cluster
    • AWS takes care of starting and stopping containers
  • Fargate
    • No need to provision the infrastructure (No EC2 instances to manage)
    • Serverless, Define the CPU and RAM you need
    • To scale, Increase the number of tasks
  • ECS anywhere
• IAM Roles
  • EC2 Instance Profile (EC2 Launch Type only)
    • Used by the ECS Agent
    • Makes API calls to ECS service
    • Send container logs to CloudWatch Logs
    • Pull Docker image from ECR
    • Reference sensitive data in Secrets Manager or SSM Parameter Store
  • ECSTask Role
    • Allow each task to have a specific role
    • Use different roles for the different ECS services
    • Task Role is defined in the task definition
• Load balancer Integrations
  • ALB is supported and works for most use cases
  • NLB recommended only for high throughput and high-performance use cases, or to pair it with AWS Private Link
  • Classic LB is supported but not recommended
• Data Volumes (EFS)
  • Mount EFS file systems onto ECS tasks
  • Works for both EC2 and Fargate launch types
  • Tasks running in any AZ will share the same data in the EFS file system
  • Fargate + EFS = Serverless
  • Use cases: persistent multi-AZ shared storage for containers
• Application Type
  • Service - A group of tasks handling long-running computing work that can be stopped and restarted. For example, a web application
  • Task - A standalone task that runs and terminates. For example, a batch job
• Scaling
  • Auto Scaling Group Scaling
    • Scale ASG based on CPU Utilization
    • Add EC2 instances over time
  • ECS Cluster Capacity Provider
    • Used to provision and scale the infrastructure for your ECS Tasks automatically
    • Capacity Provider paired with an Auto Scaling Group
    • Add EC2 instances when missing capacity (CPU, RAM, ...)

ECR

• Elastic Container Registry

EKS

• Elastic Kubernetes Service

EKS assumes role

aws eks update-kubeconfig --name <cluster-name> --role-arn arn:aws:iam::<account-id>:role/<role-name> --region ap-southeast-1

Fargate

• Serverless Container Service

App Runner