1. 程式人生 > 其它 >[SAA + SAP] 03. EC2

[SAA + SAP] 03. EC2

SAA

Security Group

SG can refer to another SG.

EC2

  • You have more access to hareware
  • Whole hardware for your EC2
  • Still share hardware with other instances
  • To start a spot instance, you need to issue a Spot Request.
  • Spot request contains the information
    • Max price
    • Num of Instances
    • Specification
    • Type: One-time | presistent
    • Time range
  • As long as current spot price < Max price, instances keep running
  • Otherwise, will stop or terminate instance in 2 mins grace period
  • Or using Spot block strategy
  • If one-time type, will terimated instances in the end.
  • Must firsts cacnel the Spot Request first, then terminate the instances, otherwise, for persisten type, it will "stop" -> "start" -> "according to spot request" -> "lanuch instances" again.
  • You cannot cancel "failed" or "closed" instances
  • Launch spot instances and on-demand instances to meet target capacity.
  • Start/stop EC2, public IP will change, use Elastic IP to avoid this problem
  • Recommended to use a random Public IP and register a DNS name to it.
  • Or use a Load Balancer are ofter a better choice.

EC2 Placement Group

  • Sometime you want to control over the EC2 insstnaces placement strategy (to achieve low-latency or aviliability)
  • That strategy can be defined using placement groups
  • You can move an instance into or out of a placement Group
    • You first need to stop the instance
    • You then need to use CLI (modify-instance-placement)
    • You can then start your instance
  • Note: Choose instance type that has Enhanced Netowrking (because they have faster networking capbility)
  • Put all EC2 in the same Rack
  • To get Low latency network
  • For big data job
  • Spread: accross AZ, limited 7 instance per AZ
  • For reduce risk of simultaneous failure
  • For Critical application
  • Partition: Each partition pan access multi AZ in same region
  • Up to 100 instances
  • For Kafka, HDFS...
  • Can attach to EC2 instances for failover
  • preserve the state in EBS
  • reduce bootstrap time compare to stop instances

EC2 Nitro

  • Underlying Platform for the next generation of EC2 instances
  • New virtualization technology
  • Allow for better performance:
    • Better networking options (enhanced networking HPC, IPv6)
    • Higher Speed EBS (Nitro is necessary fro 64,000 EBS IOPS - max 32,000 on non-Nitro)
  • Better underlying security

If you want High Speed EBS, you have to use EC2 Nitro instance.

vCPU

  • Core has Thread inside
  • You can reduce some vCPU to reduce some costs

SAP

  • Host affnity for Dedicated host, when the instance reboot, it will stay on the same host; won't jump to another hardward
  • R for RAM
  • C for Compute
  • M for Web, Medium
  • I for I/O
  • G for GPU machine learning
  • Status check for EC2 VM and system
  • RAM is not included in EC2 metrics, need to use Custom metrics
  • Focus on capacity, can mix the instance types