High Availability

The ability for a system to remain available

Think about what could cause a service to become unavailable

1. When an AZ becomes unavailable eg. data-center flooded

2. When a Region becomes unavailable eg. meteor strike

3. When an web-application becomes unresponsive eg. too much traffic

4. When an instance becomes unavailable eg. instance failure

5. When a web application becomes unresponsive due to distance in
   geographic location

The solution we need to implement in order to ensure High Availability:

1. We should run our instances in Multi-AZ, an Elastic Load Balancer can route traffic to operational AZs.

2. We should run instances in another region. We can route traffic to another Region via Route53

3. We should use Auto Scaling Groups to increase the amount of instances to meet the demand of traffic

4. We should use Auto Scaling Groups to ensure a minimum amount of instances are running and have ELB route traffic to healthy instances

5. We should use CloudFront to cache static content for faster delivery in nearby regions. We can also run our instances in nearby regions and route traffic using a geolocation policy in Route53

Scale Up vs Scale Out

When utilization increases and we are reaching capacity we can:

Scale up (Vertical Scaling)

Increasing the size of instances
- Simpler to manage.
- Lower availability (if a single instance
fails service becomes unavailable)

Scale out (Horizontal Scaling)

Adding more of the same
- More complexity to manage.
- Higher availability (if a single instance
fail it doesn't matter)

You will generally want to scale out and then up to balance complexity vs availability