EC2_01: https://chatgpt.com/c/7ddee1bf-7240-4789-913a-72dae8438d00
EC2_02 : https://chatgpt.com/c/636011e3-1ce6-4268-8089-47ca3b12c9b9

EC2 / ECC (elastic compute cloud)¶

1. Intro¶

IaaS
connect from terminal (instance must have public IP)
SSH (OS : linux,mac,W10+)
- generate SSH key-Pair
Putty ( OS: W7,8,9)

EC2 instance Connect (Amazon Linux 2) - browser session.

- aws cli installed in ec2
- run any commnd, provided IAM permission added on ec2-role
- don't run "aws configure"

get Instance metadata
http://169.254.169.254/latest/meta-data/ (169.254)
instance ID,
public IP, private IP, elastic IP, etc
...
Retrieve runtime information dynamically
AMI
ami accidental deledte --> goes to recycle bin
AMIs are bound to the Region they are created in.
So, you need to copy the AMI across Regions for disaster recovery readiness.
When the new EBS-backed AMI is copied from Region A into Region B,
- no charges for copying an AMI.However, standard storage and data transfer rates apply.
- it automatically creates a snapshot (root volumee) in Region B because AMIs are based on the underlying snapshots
- Region B will have :
- 1 AMI
- 1 snapshot (root + additional optional EBS vol)
- https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/CopyingAMIs.html

t2.micro      1     1     EBS-Only          Low to Moderate
t2.xlarge     4     16    EBS-Only          Moderate
c5d.4xlarge   16    32    1 x 400 NVMe SSD  Up to 10 Gbps     
r5.16xlarge   64    512   EBS Only          20 Gbps           
m5.8xlarge    32    128   EBS Only          10 Gbps

- instance profile

- container for a role that can be attached to an Amazon EC2 instance when launched - contain only one role, and that limit cannot be increased

2. ec2: Instance type (family,gen,size) :yellow_circle:¶

https://ec2instances.github.io/
https://instances.vantage.sh/

general purpose :¶

balanced,
eg: t2.micro

compute optimized,¶

gaming server, media, batch, ML, HPC

memory optimized¶

real-time system, large dataset processing, BI

storage optimized¶

File System, RDBMS, data-ware house, cache server, OLTP etc

3. Purchase options :yellow_circle:¶

3.1 On-Demand¶

No long-term commitment
no upfront payment
Flexible
use case: short workload
pay by
second (after the first minute) (Window and Linux)
hour (other OS)

Capacity Reservations¶

Reserve On-Demand instances capacity in a specific AZ for any duration

3.2 Reserved¶

1 & 3 years
fixed : InstanceType, scope:Region/AZ, Tenancy, OS
up to 72% discount
Payment Options – No Upfront (+), Partial Upfront (++), All Upfront (+++)
use case:
long workloads
steady-state usage applications

Convertible Reserved¶

upto 66% discount
flexible : Instance Type, scope:Region/AZ, Tenancy, OS
use case: long workloads with flexible instances types

3.3 Savings Plans¶

(1 & 3 years)
commitment to an amount of usage.
Usage beyond EC2 Savings Plans is billed at the On-Demand price
fixed : instance family & AWS region
Flexible : Instance Size, OS, Tenancy

3.4 Spot Instances¶

upto 90% discount + MOST cost-efficient
can lose instances (less reliable),
When we cancel an active spot request, it does not terminate the associated instance
2 min grace period.
use case:
Not suitable for critical jobs or databases
workloads that are resilient to failure : Batch jobs, Data Analysis, Image processing, etc
type:
one time
- if interrupted then terminated.
persistent
- has to terminate instance manually

provision : define launch pool

launch pool :
    duration/expiry            <<<
    type: one-time/persistent 
    count
    instance Types 
    cpu (min.max)    

-   define **our max spot price** , else od-price

        if max spot price  < current Spot price 
           then use it
        if max spot price  > current Spot price
            then stop/terminate instance
            with 2 min grace

terminate:

3.4.1 Spot fleet¶

collection of Spot Instances (and optionally On-Demand Instances)
on-time Instances only
define multiple launch pool
thus request various instance types
provides higher availability and fault tolerance
if spot instance is termninated, it will match with another spot instance
if none found then OD
thus maintain target capacity
strategies:
lowestPrice:
- from the pool with the lowest price
- (cost optimization, 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 highest capacity available,
- then selectthe pool with the lowest price (best choice for most workloads)

3.4.1 Spot block (deprecated) ¶

Spot Blocks are priced at the same rate as Spot Instances
but with the added benefit of no interruptions for the block duration (1-6 hr)

3.6. Dedicated Instances¶

underlying h/w not shared with other account.
May share hardware with other instances in same account

3.7 Dedicated Hosts¶

most expensive $$
book an entire physical server/racks/building.
for companies that have strong regulatory or compliance needs
control over instance placement
pay by :
physical cores
underlying network socket visibility.
Purchasing Options
On-demand – pay per second for active Dedicated Host
Reserved - 1 or 3 years (No Upfront, Partial Upfront, All Upfront)

summary¶

¶

4. hibernate¶

RAM state ( < 150 GB), gets preserved in
EBS root volume
with/without Encryption
hence boot up fast.
supported family instances : C3, C4, C5, I3, M3, M4, R3, R4, T2, T3, …
C**
M**
T**
max for 60 days
Also, Root Volume – must be EBS, encrypted.

#1
A Machine Learning research group uses a proprietary computer vision application hosted on an Amazon EC2 instance. 
Every time the instance needs to be stopped and started again, the application takes about 3 minutes to start 
as some auxiliary software programs need to be executed so that the application can function. 
The research group would like to minimize the application boostrap time whenever 
the system needs to be stopped and then started at a later point in time.

As a solutions architect, which of the following solutions would you recommend for this use-case?
- Create an Amazon Machine Image (AMI) and launch your Amazon EC2 instances from that
- Use Amazon EC2 Instance Hibernate **

¶

5. Placement group :yellow_circle:¶

cluster
low-latency group in a single AZ
close-proximity.
hence high n/w throughput
usecase : HCP
spread
spreads instances across underlying hardware in mutli-AZ
max : 7 instance/group/AZ
partition
partition-1 in AZ-1
partition-2 in AZ-1
...
partition-3 in AZ-2
...
max /restriction
- 100 instances/group
- 7 partitions/AZ
use case: hadoop, kafka, casandra

6. provision instance :yellow_circle:¶

Tenancy
host / physical building and racks
Dedicated / single-tenant (single aws account)
default/shared/multi-tenant (multiple aws account, shares same hardware)
change tenancy:
``` dedicated <==> host : can change cannot change any other, fixed once --- some rules to remember:
1. VPC tenancy take precidence over ASG's LT's tenancy
2. host > dedicated > default : cannot downgrade. ```
OS / AMI
can create AMI on ec2-i.
OS boot volume : gp* / io*
memory/RAM
cpu architecture
ARM
AMD
user-data script
bootstrap script - automate boot tasks
- install update
- install software
- ...
run only once (first launch by-default)
- can change to run everytime instance restarted
run with root user permission.
Storage
02_storage
Network-attached (EBS & EFS)
hardware (EC2 Instance Store)
Network :
04_network
ENI 00_eni+sg.md
sg: 00_eni+sg.md
Elastic load Balancer 01_ELB.md
Auto Scaling group 01_ASG.md

7. Save cost ¶

AWS Cost Explorer¶

helps you identify under-utilized Amazon EC2 instances.
- that may be downsized on an instance by instance basis within the same instance family,
understand the potential impact on your AWS bill by taking these purchace option:
- Reserved Instances
- Savings Planss

AWS Compute Optimizer¶

recommends optimal AWS Compute resources for your workloads
to reduce costs and improve performance by using machine learning
- to analyze historical utilization metrics.
- helps you choose the optimal Amazon EC2 instance types.