Public IP, Private IP, Elastic IP

PUBLIC IP :
- machine identified on www
- unique across whole web
- can be geo located
PRIVATE IP :
- machine identified on private network
- unique across private network
- two different networks can have same private IP
- only specified range of IPs can be used
- machines connect to www via NAT device + internet gateway (proxy)
ELASTIC IP :
- On stopping and restarting an EC2 instance, its public IP changes
- Elastic IP is a fixed public IPv4 IP for the instance that doesn't change until the instance is deleted
- can be attached to one instance at a time
- You can only have 5 elastic IP in your account
- DISADVANTAGE : reflects poor architectural decisions
- ALTERNATIVES :
  - using random public IP and register a DNS name to it
  - using load balancer
In general, an EC2 instance has
- a private IP -> internal AWS network
- a public IP -> www
Public IP has to be used to SSH into EC2 instance as we are not in same network to use private IP

Placement Groups

The strategy of how to place the EC2 instance within the AWS infrastructure can be defined using placement groups
3 strategies for the group :
- CLUSTER :
  - instances grouped together in a low latency hardware setup within single AZ
  - ADVANTAGE :
    - high performance
    - super low latency
    - great network speed
  - DISADVANTAGE :
    - high risk
    - one hardware fails, all instances fail
  - USE CASE :
    - big data job
    - an application that requires extremely low latency and high network throughput
- SPREAD :
  - instances spread across different hardware
  - instances' failure are isolated
  - ADVANTAGE:
    - minimized risk of simultaneous failure
    - span across multiple AZ
  - DISADVANTAGE :
    - can have only 7 EC2 instances per placement group per AZ
  - USE CASE :
    - application requiring high availability
    - CRITICAL APPLICATIONS
- PARTITION :
  - instances spread across different partitions; partitons spread across different sets of hardware within an AZ
  - instances' failure are not isolated but one partition isolated from another partition of failure
  - can have & partitions per AZ
  - ADVANTAGE :
    - span across multiple AZ in single region
    - run 100s of EC2 instances
    - each partition failure is isolated
    - EC2 instances get access to partition info as metadata
  - USE CASE :
    - HDFS
    - Hadoop
    - Cassandra
    - Kafka

represents virtual network card
a logical component in a VPC
they give EC2 instances access to the network
ENI has :
- Primary private IPv4
- one or more secondary IPv4
- one elastic IPv4 per private IPv4
- one public IPv4
- one or more security groups
- a MAC address
can create ENI independently and attach them to EC2 instances on the fly (move them) when a failover occurs
bound to a specific AZ

on starting an EC2 instance,
- OS boots and EC2 user data script runs
- OS boots up
- application starts
- caches get warmed up
this takes time
on hibernating EC2 instance,
- in-memory (RAM) state is preserved
- instance boot will be faster
- the RAM state is written to a file in root EBS volume
- root EBS volume must have enough space and must be encrypted
USE CASES :
- for long-running process
- saving RAM state
- (faster boot up) services that take time to initialize
instance RAM size must be less than 150 GB
Hibernation not more than 60 days

multilple threads run on one CPU
each thread is a vCPU
eg : m5.2xlarge
- 4 CPU
- 2 threads per CPU
- => 8 vCPUs
you may need to change no of vCPUs to:
- decrease no of CPU cores
- decrease no of threads per Core

if you need to plan in advance for capacity and make sure that you can launch instances in an AZ of a specific type around the specific timeframe then use EC2 capacity reservations.
get billed as soon as it starts
no planned end-date for reservation
specify :
- AZ
- no of instances
- instance attributes
  - type
  - tenancy
  - platform/OS

Spot Fleet is a set of Spot Instances and optionally On-demand Instances. It allows you to automatically request Spot Instances with the lowest price.