Disaster Recovery Strategies

A key motivation for adopting public clouds is the capabilities that they provide, like geographic redundancy and compute bursting. Compute bursting is a functionality that on-premises applications can use when they run out of resources or capacity; it enables them to offload to the cloud when demand increases or a failure arises.

AWS provides several building blocks for disaster recovery operations by leveraging low costs and operations. In this chapter, we will discuss the three main models that users can adapt to increase their business continuity.

The following topics will be covered in this chapter:

• Availability metrics
• Backup and restore
• Pilot light
• Warm standby
• Multi-site active-active

Technical requirements

There is no technical requirement for this chapter.

Availability metrics

High availability (HA) is a complex variable, because a deep understanding of how every component interacts with each other must be in place. Metrics come to the rescue; they provide a quantitative perspective of the system operations discipline, but in some cases, not all of the factors are correlated, and a qualitative perspective can help. This database of indicators can be used to build a model to represent the availability qualities of a system.

A system's availability can be calculated by the number of nines in the digits, representing the percentage of time a system is operating, without downtime. This is shown in the following table:

Number of nines	Uptime %	Downtime per day
1	90%	2.4 hours
2	99%	14 minutes
3	99.9%	86 seconds
4	99.99%	8.6 seconds
5	99.999%	.86 seconds

...

Backup and restore

This strategy consists of the active replication of all of the primary assets that will be needed to recover service on the cloud. This can include files, databases, disks, and virtual images, to name a few.

The following diagram shows the overall solution for a backup and restore strategy:

Preparation phase

In the preparation phase, a golden AMI needs to be created, with all of the required dependencies to run the application. You also have to allocate an Elastic IP that will be used to route production traffic to the disaster instance. This instance will be created when the failure is notified.

EBS volumes can be created on demand, with the latest snapshot; services like AWS Storage Gateway replicate...

Pilot light

Pilot Light uses the analogy of a gas heater, in which the standby system is always on, ready to ignite the complete system when the failure occurs. Primary data must be frequently replicated to the cloud. The premises database layer needs to be in sync with a standby replica in the RDS.

The following diagram shows the overall solution for a backup and restore strategy:

The preparation phase

Database synchronization must be configured from the source (production) to the pilot light environment (warm standby). To achieve this, you can use the vendor standard replication scheme to replicate the data layer transactions asynchronously. The Database Migration Service can also be used, leveraging the high throughput...

Warm standby

In this scenario, the warm standby is always running and fully functional, but with the minimal amount of resources, to keep the costs at a minimum. This environment can be used for testing and quality assurance. It is illustrated in the following diagram:

The preparation phase

To prepare, create the required golden AMIs, with all of the dependencies and code built in, so the service can fail over at any time, as a result of a disruption in the active site (on premises). Prepare a replication instance that is constantly updated with the latest primary site transactions.

In the case of a disaster

...

Multi-site active-active

The multi-site active-active strategy provides strong advantages, as compared to other DR scenarios, because a redundant service architecture is running in the cloud. In this scenario, the primary service (on-premises) can failover at any time, and can cope with the production traffic. A strong point of multi-site active-active is that traffic can be split for AB testing, game day exercises, and custom traffic distribution. Multiple regions can be used to achieve golden availability metrics with five nines:

The preparation phase

To prepare, migrate the business-essential applications, and perform database synchronization. This replication can be done synchronously or asynchronously, depending on the...

Summary

In this chapter, we learned about the main patterns used in DR strategies by using the cloud. We learned to implement successful backup and restore, use a pilot light, and multi-site active-active scenarios.

We also learned how to implement a full DR exercise in a hybrid environment.

Further reading

The implementation details in this chapter represent an opinionated view of how to recover using managed services, and do not represent the standard perspective of AWS. Use the implementation details as guidelines only. For more information, read the white paper, Using Amazon Web Services for Disaster Recovery, located at http://bit.ly/s2QC6P.