[AWS] Architecture Patterns - Compute

High availability and elastic scalbility for web servers?

Use EC2 Auto Scaling and An application load balancer across multi AZs.

 

Low-Latency connections over UDP to a pool of instances running a gaming application?

Use a Network Load Balancer with a UDP listener

 

Clients need to whitelist static IP addresses for a highly available load balanced application in an AWS Region?

Use an NLB and create static IP addresses in each AZ.

 

Application On EC2 in an Auto Scaling group requires disaster recovery across Regions?

Create ASG in a second region iwth the capacity set to 0. Take snapshots and copy them across Regions (Lambda or DLM).

 

Application on EC2 must scale in larger increments if a big increase in traffic occurs, compared to small increases in traffic?

Use Auto scaling with a Step Scaling policy and configure a larger capacity increase.

 

Need to scale EC2 instances behind an ALB based on the number of requests completed by each instance?

Configure a target tracking policy using the ALBRequestCountPerTarget metric.

 

Need to run a large batch computing job at the lowest cost. Must be managed. Nodes can pick up where others left off in case of interruption?

Use a managed AWS Batch job and use EC2 spot instances.

 

A tightly coupled High Performance Computing (HPC) workload requires low-latency between nodes and optimum netowork perofrmance?

Launch EC2 instances in a signle AZ in a cluster placement group and use an Elastic Fabric Adapter (EFA)

 

LOB Application receives weekly burst of traffic and must scale for short periods - need the most cost-effective solution?

Use reserved instances for minimum required workload and then use Spot instances for the bursts in traffic.

 

Application must startup quickly when launched by ASG but requries app dependencies and code to be installed?

Create an AMI that includes the application dependencies and code.

 

Application runs on EC2 behind an ALB. Once authenticated users should not need to reauthenticate if an instance fails?

Enable sticky session for the target group or alternatively use a session store such as DynamoDB