I M P L E M E N T I N G H I G H AVA I L A B I L I T Y
W I T H AW S
CONTENTS
AWS Regions and Availability Zones
Auto-Scaling Group and Load Balancer
Disaster Recovery Strategies
AW S R E G I O N S A N D AVA I L A B I L I T Y
ZONES
AW S R E G I O N S
Physical locations where AWS resources (such as servers) are actually
located
Each region is a geographically separate area
Each AWS Region is a separate geographic area. Each AWS Region has
multiple, isolated locations known as Availability Zones.
Each region has different number of Availability Zones.
We can deploy redundant resources (EC2, databases, etc.) in multiple
Availability Zones as a fail-safe in case one of the AZ is compromised/go
out of service.
For the highest availability possible, we might even deploy backup
resources in multiple regions, in case one of the regions go out of service
(e.g. in a nation-wide disaster scenario). Drawbacks include possible
higher latency and higher cost.
Most S3 services are stored in multiple Azs by default (notable exceptions
AUTO-SCALING (COMPUTE/EC2)
Scale the EC2 instances horizontally (adding new instances instead of
increasing RAM and/or CPU capacity)
We need to create an AMI (templates) of our main EC2 instance
containing the application so AWS can use it when launching other EC2
instances.
We can set dynamic scaling policy based on resource usage (e.g. when
CPU utilization is more than 50% we can automatically launch another
EC2 instance, and when it reaches less than 20%, we can remove an
instance to save cost)
We can also set the scaling policy based on a time schedule for more
predictable workloads
We can configure autoscaling to use multiple AZs for higher availability
(not regions though)
AWS will try to maintain the minimum capacity by automatically
replacing unhealthy instances.
AUTO-SCALING (OTHERS)
Autoscaling for some resources, such as databases, might be not as
simple as autoscaling for compute resources.
For example, AWS might use the read replica system for reducing the
load of a database. If required, read replica might also be promoted to a
main database.
Cross-regions read replica is available for certain regions (but not all)
E L AS T I C LOAD B AL AN C E R
Load balancer automatically distributes workload to several AWS
resources.
Application Load Balancer works on Application Layer (OSI layer 7), which
means that this operates on HTTP/HTTPS protocols. This could serve as a
single point of contact for clients.
There are other type of ELB, such as Network Load Balancer which
operates on TCP/UDP protocol (OSI layer 4)
We can create listener rule and distributes them to the appropriate target
groups of resources (auto scaling groups).
The default algorithm is round robin.
For multi-regions configuration, we can set a DNS service in front of the
load balancers.
E L AS T I C LOAD B AL AN C E R
Load balancer automatically distributes workload to several AWS
resources.
Application Load Balancer works on Application Layer (OSI layer 7), which
means that this operates on HTTP/HTTPS protocols. This could serve as a
single point of contact for clients.
There are other type of ELB, such as Network Load Balancer which
operates on TCP/UDP protocol (OSI layer 4)
We can create listener rule and distributes them to the appropriate target
groups of resources (auto scaling groups).
The default algorithm is round robin.
For multi-regions configuration, we can set a DNS service in front of the
load balancers.
AW S D I S A S T E R R E C O V E RY S T RAT E G I E S
AWS provides four strategies for disaster recovery based on cost, RTO,
and RPO requirements.
RTO: the time a system experiences downtime. RPO: the maximum
amount of data the org can afford losing (measured in time). RPO is the
time between the last full data backup.
Backup and Restore (cheapest but highest RPO and RTO)
Pilot Light (more expensive than Backup and Restore, provides better
RPO and RTO)
Warm Standby (more expensive than Pilot Light, even better RPO and
RTO)
Active/Active (very costly, but virtually zero RPO and RTO)
BACKUP AND RESTORE
P I LOT L I G H T
WA R M S TA N D BY
ACTIVE/ACTIVE
T H A N K YO U
