Elastic Compute Cloud¶

TLDR¶

EC2 is a combination of a virtual machines and hardware capacity attached to that VM in AWS. It is one of the oldest services and well integrated into most other services.

Virtual Machines (EC2)¶

Configuration Options¶

• OS (Win, Linux, MacOS)
• CPU
• RAM
• Network attached storage (EBS & EFS)
• can have multiple EBS attached storage
• Hardware attached (InstanceStore)
• Network Card (Speed, Ip)
• Firewall (SecurityGroup)
• Bootstrap script (User Data)

User Data¶

• Is a script is run once on the first start of the instance
• Script is run using root privileges

Use Cases¶

• Installing updates
• Installing software
• Download files from the internet

Instantiating EC2 Instances quickly¶

• Use Golden AMI(pre-configured AMI with all dependencies installed)
• Boostraping using User data : For dynamic config, user data scripts
• Hybrid: mix Golden AMI and User Data ElasticBeanstalk

Instance Types¶

• https://instances.vantage.sh/

Type Definition¶

• Split by family
• In family types are defined by Class.Generation.Size

Families¶

General Purpose¶

• Balance between, computing, memory and network speed

Compute Optimized¶

• e.g instance name c6g => (where c stands for cpu)

Use cases¶

• High cpu power
• batch processing workloads
• media transcoding
• high performance web servers
• high perfomance computin (HPC)
• Scientific modeling & machine learning
• dedicated gaming servers

Memory Optimized¶

• e.g instance name r6g => (where R stands for Ram)
• High amount of RAM/Memory

Use cases¶

• high performance relational / non relational databases
• distributed web scale cache stores
• In-memory databases optimized for BI
• Applications performing real-time processing of big unstructured data
• Process large data sets in memory
• Cache stores

Storage Optimzied¶

• High amount of Disk IOPs
• Great for storage-intensive tasks that require high, sequential read and write acccess to large datasets on local storage.

Use cases¶

• High frequency online transaction processing systems (OLTP)
• Relational & NoSQL databases
• Cache for in-memory databases (e.g redis)
• Data warehousing apps
• Distributed file systems.
• Alot of Datasets in local storage

Purcharsing Options¶

• On-Demand Instances - short workload, predictable pricing , pay by second.
• Reserverd ( 1 & 3 years)
  • Reserved Instances - long workloads
  • Convertible Reserved Instances - long workloads with flexible instances
• Saving Plans (1 & 3 years)
• Spot Instances - short workloads, cheap, can lose instances (less reliable)
• Dedicated hosts - book an entire physical server, control instance placement
• Dedicated Instances - no other customer will share your hardware.
• Capacity Reservations - reserve capacity in a specific AZ for any duration.

On Demand $$$$¶

best for shor-term and un-interrupted workloads, where you can't predict how the app will behave.

• Pay for what you use:
  • Linux or windows - billing per second, after the first minute.
  • all other OS - billing per hour.
• has the highest cost but no upfront payment
• no long-term commitment
• Use for short workload

Reserved¶

Default¶

Recommended for steady-stage usage aps e.g database

• Up to 72% discount compared to on-demand.
• Cheaper if payed upfront & Cheaper than on demand.
• Need to specify type os, region, tenacy
• Reservation Period - 1 year (+ discount) or 3 years (+++)
• Payment Options - No Upfront (+), Partial Upfront (++), All Upfront (+++)
• Reserved Instance Scope - Regional or Zonal (reserve in zone and AZ)
• You can buy or sell in the Reserved Instance marketplace.

Convertiable¶

• Up to 66% discount
• More expensive than default
• Can change instance type, os, region, tenacy

SavingsPlans¶

• 1 Year OR 3 Years
• Get a discount based on long-term usage (up to 72% - samge as RIs)
• Commit to a certian type of usage ($10/hour for 1 or 3 years)
• usage beyond commit is billed on demand price $$$$
• Locked to instance family and region
• Commit to an amount of usage (cpu, mem, disk)
• Flexible (can change)
  • Instance Size (m5.xlarge, m5.2xlarge)
  • OS (Linux, windows)
  • Tenancy (Host, dedicated, default)

Spot Instances¶

Useful for workloads that Fault Tolerant - discount of up to 90% vs on-demand. - Cheapest of all types - Instances can be lose at any point in time. - MOST cost-efficient instances in AWS. - Define max price you are willing to pay: - if you are overbid your instance is gone. - at any time Can lose instance

Use Cases¶

• Short workloads
• batch jobs
• Suited for resiliant jobs
• Image processing
• Data analysis
• Distributed workloads (these are resiliant by definition)
• workloads with a flexible start and end time.
• not suitable for critical jobs or databases.

Dedicated Hosts¶

• Rent an entire physical server, control placement of that server
• allows address compliance requirements and use your existing server software licenses (per-socket , per-core, per VM software licenses)
• useful for software that have complicated licensing model (BYOL- bring your own license)
• can use on demand or reserved.
• most expensive
• can convert to dedicated instance

Use Cases¶

• Complicance
• Server bound software licences

Dedicated Instances¶

• No other customer will share the hardware used by you
• Hardeware may be shared within same account
• No control over instance placement (can move hardware after Stop /start)
• can convert to dedicated host

Capacity Reservations¶

• Reserve a capacity in a AZ for duration (on demand instance)
• No time commitment
• No billig discount
• Combine with Saving plan and Reserved Instances for same AZ to save money
• Charged at on-demand rate whether you run instances or not.

#### Use Cases - short term uninterrupted workload in a specific az

Spot Instance¶

• Request for an instance and the instance running itself is diffrent
• spot request can either be one time or persistent
  • persistent request will continue supply of instances until a vlid time and desired number of instances are fullfilled.
• define max spot price and get instance while current spot price < max
  • hourly spot price varies based on offer and capacity
  • if current spot price > max_defined_price - you can either stop or terminate your instance with a 2 minutes grace period.
• If you terminate your instance but not the request the instance will be relaunched
• can only cancel a spot request , if the request is either in open , active or disabled.
• If you cancel your request your instance will still be running

Spot Blocks (Deprecated)¶

• Duration window of your Spot instance
• 1 to 6 hours
• designed to not be interrupted
• still not garanteed

Fleet¶

• define launch pools : instance type , OS , AZ.
• Define a budget and target capacity
• multiple launch pools
• Spot fleet stops launching instances when reaching capacity or max costs.
• allows us to auto request spost instances with the lowest price.

Strategies¶

• Lowest price, launch from lowest price pool
• Diversified, launch from all pools (resiliance)
• Capacity optimized, launch from pool with strongest capacity settings
• price capacity optimized (recommended) pools with highest capacity available, then select the pool with the lowest price(best choice for most workloads).

Elastic Ip¶

• public static ip
• A static ip which can be attached to Elastic Network Interfaces and other services
• Used to mask a fail by keeping ip but changing machine
• Up to 5 per account
• Can request more than 5 from aws

Placement Groups¶

Placement strategy for a group ec2 instances

Spread¶

• Multi AZ
• Different hardware
• Limited to 7 instances per AZ per placement group
• Use for critical apps and High Availability

use case¶

• app that needs to maximize High Availability
• critical apps where each instance must be isolated from failure from each other.

Cluster¶

• all instaces are on the Same rack on same AZ.
• high-performance networking
• Rack fail -> all EC2s fail
• Fast internal connection from ec2 to ec2
• Used for low latency and big data jobs with deadline
• Great network (10 Gbps bandwith between instances)

use case¶

• big data job that needs to complete fast
• app that needs extremely latency and high network throughput

Partition¶

• Multi AZ.
• Multi rack but same rack per partition
• Up to 7 partitions per AZ.
• spread instances across many partitions within an AZ
  • each partition uses a different set of racks.
• scale to 100 EC2 instances per group (Hadoop, Cassandra , Kafka).
• failure is isolated at partition level.

Use Case¶

• big data application which are partition aware (HDFS,HBase, Cassandra, Kafka)

ENI Elastic Network Interfaces¶

• Virtual Network Card
• Can be used not only for ec2
• Eth0 = primary ENI
  • 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
• you can create ENI indepently and attach them on the fly (move them) on EC2 instances for failover.
• Can add multiple for multiple private ips
• Can attach elastic ip
• Can attach mac address
• Can attach and detach from instance
• Bound by AZ

EFA Elastic Fabric Adapter¶

• use to accelerate high performance computing by providing os bypacc hardware interface
• cannot be used for windows

ENA Elastic Network Adapter¶

• use for high performance networking capabilitiies
• slower than efa

EC2 Hibernate¶

• Standby for EC2, ram is saved
• Faster startup
• Root EBS must be encrypted
• Used for breaks in very long running tasks or if an instances takes very long to startup
• Can last no longer than 60 days
• Must be enabled before launch

EC2 Instance Recover¶

• Creates a new EC2 instance which is identical to the previous ec2
• recovers private and public ip, metadat, elastic ip, instandce id
• data in memory is lost

Limits¶

• by default your account has a maximum limit for ec2 instances based on the total vcpu used, you can submit a request to increase that limit