Ref: https://learn.cantrill.io/courses/1820301/lectures/41301439 and https://learn.cantrill.io/courses/1820301/lectures/41301440

• Summary:
• Contents:

Traditional Load Balancer (LB)

• 🔧 Server that accepts connections from clients, then distributes those connections in a balanced way across any registered backend resources
• ❗ Clients are abstracted away from physical infrastructure!
  • Backend resources can fail or scale seamlessly, without disrupting the client's experience

AWS Elastic Load Balancer (ELB) - Architecture

Diagram: https://github.com/acantril/aws-sa-associate-saac03/blob/main/1500-HA_and_SCALING/00_LEARNINGAIDS/ELB-1.png

• 🔧 ELB = AWS-managed LB in your VPC
  • Deploys nodes in at least 2 AZs → ELBs are HA across AZs, and can scale
  • Backend resources are registered in the ELB's Target Group (TG)
    • Can be EC2 instances, an EC2 ASG, Lambda functions…
  • 💡 AWS likes to put the name “Elastic” in front of many of their products, emphasizing the elastic capability of the cloud
  • ‼️ Officially part of Amazon EC2! → configured in EC2 console UI
    • Historical reasons: before, only EC2 instances accepted as backend resources
    • Nowadays EC2 instances, ECS clusters, Lambda functions, etc can be targets (backend resources) of an ELB
• Configuration:
  • IPv4-only OR dual-stack (IPv4 + IPv6)
    • DNS record(s) (A or AAAA record(s)) which resolve(s) to ELB nodes
    • Any connections to the DNS name are distributed equally through ELB nodes
  • AZs to use (specifically, one subnet per AZ, in at least 2 AZs)
    • 1+ Nodes placed in each chosen AZ subnet
      • Nodes can be replaced if failure → HA
      • Nodes can scale → more nodes created if more load coming into the ELB
  • Internet-facing ELB OR internal ELB
    • internet-facing ELB → nodes have both private and public IPs
    • internal ELB → nodes only have private IPs → generally used internally to separate different app tiers
  • Listener: controls WHAT the LB does (what it listens to and how it reacts)
    • Accepts traffic on a port & protocol
    • Communicates with targets (backend resources) on a port & protocol
• Once connection with client established, ELB connects with compute resources (EC2 instances, ECS clusters…)
  • ‼️ Backend resources can be public or private! REGARDLESS of whether the ELB is internet-facing or internal!!
    • Backend instance IPs are different from ELB IPs, subnets can also differ
• ❗ ELBs need 8+ free IPs per subnet AND a /27 or larger subnet to allow scaling
  • ‼️ Strictly speaking, a /28 subnet works!
    • 💡 That's 16 IPs (minus the 5 that are AWS reserved, that leaves 11 free IPs…)
    • …HOWEVER! AWS suggests a /27 or larger subnet so that it can scale!
    • 💡 Default to thinking of /27 as the minimum size, but if in the exam they give /28 as an option and not /27, then /28 is the right answer!

AWS ELB - Abstraction of Infrastructure

Diagram: https://github.com/acantril/aws-sa-associate-saac03/blob/main/1500-HA_and_SCALING/00_LEARNINGAIDS/ELB-2.png

• Without ELBs, clients connect to servers/instances directly → tight coupling between tiers
  • 👎 If instances fail or scale, the client experience is be disrupted
• With ELBs, there is abstraction between tiers → loose coupling between tiers
  • 👍 Instances can fail or scale, clients don't notice

AWS ELB - Cross-Zone Balancing